357 82 2MB
English Pages 377 Year 2007
Several Complex Variables and Integral Formulas
This page intentionally left blank
Several Complex Variables and Integral Formulas
Kenzo Adachi Nagasaki University, Japan
World Scientific NEW JERSEY
•
LONDON
•
SINGAPORE
•
BEIJING
•
SHANGHAI
•
HONG KONG
•
TA I P E I
•
CHENNAI
Published by World Scientific Publishing Co. Pte. Ltd. 5 Toh Tuck Link, Singapore 596224 USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.
SEVERAL COMPLEX VARIABLES AND INTEGRAL FORMULAS Copyright © 2007 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN-13 978-981-270-574-7 ISBN-10 981-270-574-0
Printed in Singapore.
To my family Machiko, Hidehiko and Yuko
This page intentionally left blank
Preface
The aim of this book is to study some important results obtained in the last 50 years in the function theory of several complex variables that are mainly concerned with the extension of holomorphic functions from submanifolds of pseudoconvex domains and estimates for solutions of the ∂¯ problem in pseudoconvex domains. This book is divided into five chapters. In Chapter 1 we recall the elementary theory of functions of several complex variables. We prove that every domain of holomorphy is a pseudoconvex open set. Moreover, we give the proof of the Hartogs theorem which means that a separately analytic function is analytic. In Chapter 2 we deal with L2 estimates for the ∂¯ problem in pseudoormander. As an application, we give the convex domains in Cn due to H¨ affirmative answer for the Levi problem. Moreover, we prove the OhsawaTakegoshi extension theorem by following the method of Jarnicki-Pflug. In Chapter 3 we construct integral formulas for differential forms on bounded domains in Cn with smooth boundary, that is, the BochnerMartinelli formula, the Koppelman formula, the Leray formula and the Koppelman-Leray formula are derived. Using the integral formula, we prove H¨ older estimates for the ∂¯ problem in strictly pseudoconvex domains with smooth boundary. Moreover, we prove bounded and continuous extensions of holomorphic functions from submanifolds of strictly pseudoconvex domains with smooth boundary which were proved by Henkin in 1972. We also prove H p and C k extensions. Finally, we prove Fefferman’s mapping theorem by following the method of Range. In Chapter 4 we discuss the Berndtsson-Andersson formula and the Berndtsson formula. As an application of the Berndtsson-Andersson formula, we give Lp estimates for solutions of the ∂¯ problem in strictly pseudo-
vii
viii
Several Complex Variables and Integral Formulas
convex domains in Cn with smooth boundary. Using the Berndtsson formula, we give counterexamples of Lp (p > 2) extensions of holomorphic functions. Finally, we introduce an integral formula which was used by Diederich-Mazzilli to prove bounded extensions of holomorphic functions from affine linear submanifolds of a smooth convex domain of finite type. Chapter 5 is devoted to the study of classical fundamental theorems in the function theory of several complex variables some of which are used to prove theorems in the previous chapters. Appendix A is concerned with the compact operator theory in Banach spaces which is used to prove Fefferman’s mapping theorem. In Appendix B we give solutions to the Exercises. I am grateful to Saburou Saitoh who suggested to me the publication of this book. I am also grateful to Heinrich GW Begehr who suggested that World Scientific might be interested in publishing this book. I would like to express my sincere gratitude to Joji Kajiwara, Professor Emeritus at the Kyushu University, who introduced me to the function theory of several complex variables when I was a student at the Kyushu University, and to Morisuke Hasumi, Professor Emeritus at the Ibaraki University, who introduced me to the theory of function algebras when I was studying at the Ibaraki University. Finally, I want to express my thanks to Ms Zhang Ji, Ms Kwong Lai Fun and the staff of World Scientific for their help and cooperation. Kenzo Adachi
Contents
Preface
vii
1. Pseudoconvexity and Plurisubharmonicity
1
1.1 The Hartogs Theorem . . . . . . . . . . . . . . . . . . . . . 1.2 Characterizations of Pseudoconvexity . . . . . . . . . . . . . 2. The ∂¯ Problem in Pseudoconvex Domains
1 19 47
2.1 The Weighted L2 Space . . . . . . . . . . . . . . . . . . . . 47 2.2 L2 Estimates in Pseudoconvex Domains . . . . . . . . . . . 53 2.3 The Ohsawa-Takegoshi Extension Theorem . . . . . . . . . 100 3. Integral Formulas for Strictly Pseudoconvex Domains 3.1 3.2 3.3 3.4 3.5 3.6
The Homotopy Formula . . . . . . . H¨ older Estimates for the ∂¯ Problem Bounded and Continuous Extensions H p and C k Extensions . . . . . . . . The Bergman Kernel . . . . . . . . . Fefferman’s Mapping Theorem . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
117 . . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
4. Integral Formulas with Weight Factors 4.1 4.2 4.3 4.4 4.5
The Berndtsson-Andersson Formula . . . . . . . . . . . . Lp Estimates for the ∂¯ Problem . . . . . . . . . . . . . . . The Berndtsson Formula . . . . . . . . . . . . . . . . . . . Counterexamples for Lp (p > 2) Extensions . . . . . . . . Bounded Extensions by Means of the Berndtsson Formula ix
117 143 152 183 196 210 245
. . . . .
245 255 264 270 281
Several Complex Variables and Integral Formulas
x
5. The Classical Theory in Several Complex Variables 5.1 5.2 5.3 5.4
The Poincar´e Theorem . . . . . . . . . The Weierstrass Preparation Theorem Oka’s Fundamental Theorem . . . . . The Cousin Problem . . . . . . . . . .
. . . .
. . . .
. . . .
. . . .
291 . . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
291 298 307 326
Appendix A Compact Operators
331
Appendix B Solutions to the Exercises
343
Bibliography
359
Index
365
Chapter 1
Pseudoconvexity and Plurisubharmonicity
In this chapter we study the properties of holomorphic functions of several complex variables and plurisubharmonic functions. We define the domain of holomorphy and the pseudoconvex open set, and we prove that every domain of holomorphy is pseudoconvex, but the converse (Levi’s problem) is left to 2.2. 1.1
The Hartogs Theorem
Definition 1.1 Let f = u + iv : Ω → C be a C 1 function in an open set Ω ⊂ Cn . For zj = xj + iyj , j = 1, · · · , n, define ∂f 1 ∂f 1 ∂f = + ∂zj 2 ∂xj i ∂yj 1 ∂u i ∂v ∂v ∂u + − = + 2 ∂xj ∂yj 2 ∂xj ∂yj and
1 ∂f 1 ∂f ∂f = − ∂ z¯j 2 ∂xj i ∂yj 1 ∂u ∂v ∂u i ∂v = − + + . 2 ∂xj ∂yj 2 ∂xj ∂yj
By definition ∂f ∂ f¯ = , ∂zj ∂ z¯j
∂f ∂ f¯ = . ∂ z¯j ∂zj
Lemma 1.1 Let Ω ⊂ Cn and G ⊂ Cm be open sets and let f : Ω → G and g : G → C be of class C k for k = 0, 1, · · · , ∞. Then, g ◦ f : Ω → C is 1
Several Complex Variables and Integral Formulas
2
of class C k . Moreover, if we write f (z) = (f1 (z), · · · , fm (z)), then m ∂g ∂ ∂fk ∂g ∂ f¯k (g ◦ f )(z) = (f (z)) (z) + (f (z)) (z) ∂zj ∂wk ∂zj ∂w ¯k ∂zj
(1.1)
k=1
and m
∂ (g ◦ f )(z) = ∂ z¯j
k=1
∂g ∂fk ∂g ∂ f¯k (f (z)) (z) + (f (z)) (z) . ∂wk ∂ z¯j ∂w ¯k ∂ z¯j
(1.2)
Proof. We prove (1.1) in case n = m = k = 1. Let f (z) = α(x, y) + iβ(x, y) and w = u + iv. Then we have 1 ∂ ∂ ∂ (g ◦ f )(z) = −i g(α(x, y), β(x, y)) ∂z 2 ∂x ∂y 1 ∂g ∂α ∂g ∂β 1 ∂g ∂α ∂g ∂β + + = + . 2 ∂u ∂x ∂v ∂x 2i ∂u ∂y ∂v ∂y Then (1.1) follows from the equalities ∂ ∂ ∂ = + , ∂x ∂z ∂ z¯
∂ =i ∂y
∂ ∂ − , ∂z ∂ z¯
∂ ∂ ∂ = + , ∂u ∂w ∂ w ¯
∂ =i ∂v
∂ ∂ − ∂w ∂ w ¯
(1.2) is proved similarly.
.
Theorem 1.1 Let Ω be a bounded open set in C and let ∂Ω consist of finite C 1 Jordan curves. For u ∈ C 1 (Ω) and z ∈ Ω, we have ∂u (ζ) u(ζ) 1 ∂ z¯ dζ + dζ ∧ dζ¯ . u(z) = 2πi Ω ζ −z ∂Ω ζ − z Proof.
¯ We fix z ∈ Ω. For ζ ∈ Ω\{z} we have
¯ u(ζ)dζ ∂u(ζ) ∧ dζ dζ . = ζ −z ζ −z
For any sufficiently small ε > 0, we set Ωε = {ζ ∈ Ω | |ζ − z| > ε}. It follows from Stokes’ theorem that ¯ 1 1 u(ζ)dζ u(ζ)dζ ∂u(ζ) ∧ dζ 1 = − . 2πi |ζ−z|=ε ζ − z 2πi ζ∈∂Ω ζ − z 2πi ζ −z Ωε We have the desired equality by letting ε → 0.
Pseudoconvexity and Plurisubharmonicity
3
Definition 1.2 Let Ω be an open set in Rn . We denote by D(Ω) (or Cc∞ (Ω)) the set of all C ∞ functions f in Ω whose support supp(f ) is a compact subset of Ω. Theorem 1.2 Let Ω be a bounded open set in the complex plane and let K ⊂ Ω be compact. Then for any open set ω in Ω satisfying K ⊂ ω, there exist constants Cj , j = 0, 1, · · · , such that sup |f (j) (z)| ≤ Cj f L1 (ω)
z∈K
for every holomorphic function f in Ω. Proof. Let K ′ be a compact set such that K ⊂ K ′ ◦ ⊂ K ′ ⊂ ω. Choose a function ψ ∈ Cc∞ (ω) with the properties that ψ = 1 in K ′ . By Theorem 1.1, we have ∂(ψf ) 1 1 dζ ∧ dζ¯ (ζ) (ψf )(z) = ¯ 2πi ζ −z ∂ζ ω = Since
∂ψ ∂ ζ¯
1 2πi
f (ζ) ∂ψ ¯ dζ ∧ dζ. (ζ) ¯ ζ −z ∂ζ
ω
= 0 in K ′ , we have f (z) =
1 2πi
ω\K ′
f (ζ) ∂ψ dζ ∧ dζ¯ (ζ) ζ −z ∂ ζ¯
for z ∈ K. By differentiating j times with respect to z, we obtain f (ζ) ∂ψ j! ¯ f (j) (z) = (ζ) dζ ∧ dζ. ¯ 2πi ω\K ′ ∂ ζ (ζ − z)k+1 If z ∈ K, ζ ∈ ω\K ′ , then there exists a constant C > 0 such that |z−ζ| ≥ C. Hence there exists a constantC1 > 0 such that |f (j) (z)| ≤ C1 |f (ζ)|dxdy (ζ = x + iy), ω\K ′
which gives the desired inequality.
Definition 1.3 Let Ω be an open set in C. Then u : Ω → R ∪ {−∞} is called subharmonic in Ω if (1) u is upper semicontinuous in Ω, that is, {z ∈ Ω | u(z) < s} is an open set for any real number s.
Several Complex Variables and Integral Formulas
4
(2) For any compact set K ⊂ Ω and any continuous function h on K which is harmonic in the interior of K, u satisfies the following properties: u(z) ≤ h(z) (z ∈ ∂K) =⇒ u(z) ≤ h(z) (z ∈ K). Definition 1.4 Let u : Ω → R be a C 2 function in an open set Ω ⊂ C. We say that u is strictly subharmonic in Ω if ∂2u (z) > 0 ∂z∂ z¯
(z ∈ Ω).
Theorem 1.3 Let Ω be an open set in C. Then a real-valued function u ∈ C 2 (Ω) is subharmonic in Ω if and only if ∂2u (z) ≥ 0 ∂z∂ z¯ Proof.
(z ∈ Ω).
Let a = α + iβ ∈ Ω. For r with 0 < r < dist(a, ∂Ω), define A(r) =
1 2π
2π
u(a + reiθ )dθ.
0
Then we have dA(r) 1 = dr 2π =
= = = 2
∂ u If ∂z∂ z¯ (z) ≥ 0, then Therefore, we obtain
2π
d u(a + reiθ )dθ dr 0 2π ∂u 1 (α + r cos θ, β + r sin θ) cos θ 2π 0 ∂x ∂u + (α + r cos θ, β + r sin θ) sin θ dθ ∂y ∂u ∂u 1 dy − dx 2πr |z−a|=r ∂x ∂y 2 1 ∂ u ∂2u dxdy + 2πr ∂x2 ∂y 2 |z−a|≤r 2 ∂2u dxdy. πr ¯ |z−a|≤r ∂z∂ z dA(r) dr
≥ 0. Hence A(r) is monotonically increasing.
1 u(a) = A(0) ≤ A(r) = 2π
0
2π
u(a + reiθ )dθ,
Pseudoconvexity and Plurisubharmonicity
5
which means that u is subharmonic. Conversely, suppose u is subharmonic. ∂2u Suppose there exists a point a satisfying ∂z∂ z¯ (a) < 0. For any sufficiently ∂2u small r > 0, if |z − a| ≤ r, then ∂z∂ z¯ (z) < 0, which implies that 2 ∂2u dA(r) = dxdy < 0. dr πr ¯ |z−a|≤r ∂z∂ z Since A(r) is strictly monotonically decreasing, we have 2π 1 u(a) > u(a + reiθ )dθ, 2π 0 which is a contradiction. Thus, we have Definition 1.5
∂2u ∂z∂ z¯ (z)
≥ 0.
For a ∈ C and r > 0, define B(a, r) := {z ∈ C | |z − a| < r}.
The closure of B(a, r) is denoted by B(a, r). Theorem 1.4 Let u be a continuous real-valued function on ∂B(0, R). For z = reiθ ∈ B(0, R), define 2π 1 R2 − r 2 U (z) = dϕ. (1.3) u(Reiϕ ) 2 2π 0 R − 2Rr cos(ϕ − θ) + r2 Then U is harmonic in B(0, R). Moreover, if we define U (z) = u(z) for z ∈ ∂B(0, R), then U is continuous in B(0, R). The right side of (1.3) is called the Poisson integral. Proof.
For |z| < R, define f (z) =
1 2π
0
2π
u(Reiϕ )
Reiϕ + z dϕ. Reiϕ − z
iϕ
For ζ = Re , we have ∞ n z Reiϕ + z . =1+2 iϕ Re − z ζ n=1
Since the right side of the above equality converges uniformly on |ζ| = R, we obtain 2π ∞ 2π 1 u(Reiϕ ) 1 iϕ f (z) = dϕ z n . u(Re )dϕ + 2π 0 π n=1 0 (Reiϕ )n
Several Complex Variables and Integral Formulas
6
Therefore, f is holomorphic in B(0, R). On the other hand we have iϕ 2π Re + z 1 iϕ u(Re )Re Re f (z) = dϕ 2π 0 Reiϕ − z 2π R2 − r 2 1 dϕ u(Reiϕ ) 2 = 2π 0 R − 2Rr cos(ϕ − θ) + r2 = U (z). Hence, U is harmonic in B(0, R). Next we fix a point ζ0 = Reiϕ0 . For ε > 0, there exists δ > 0 such that if ζ = Reϕ , |ϕ − ϕ0 | < δ, then |u(ζ) − u(ζ0 )| < ε. We can choose ρ > 0 so small that if |z| < R and |z − ζ0 | < ρ, then there exists a constant c > 0 such that R2 − r2 < εδ 2 ,
|Reiϕ − z| > cδ (ϕ0 + δ ≤ ϕ ≤ ϕ0 − δ + 2π).
We set M = max |u(z)|. |z|=R
Then we have |U (z) − U (ζ0 )| = |U (z) − u(ζ0 )| iϕ 2π 1 Re + z = (u(Reiϕ ) − u(ζ0 ))Re dϕ iϕ 2π 0 Re − z 2π 2 1 R − r2 iϕ = dϕ (u(Re ) − u(ζ0 )) 2π 0 |Reiϕ − z|2 1 ϕ0 +δ R2 − r 2 iϕ ≤ dϕ (u(Re ) − u(ζ0 )) 2π ϕ0 −δ |Reiϕ − z|2 1 ϕ0 −δ+2π 2 2 − r R dϕ (u(Reiϕ ) − u(ζ0 )) + 2π ϕ0 +δ |Reiϕ − z|2
ϕ0 +δ
M R2 − r 2 dϕ + iϕ 2 |Re − z| π ϕ0 −δ 2M ε 2M ≤ε+ 2 =ε 1+ 2 . c c
≤
1 2π
ε
Hence, U is continuous in B(0, R).
ϕ0 −δ+2π
ϕ0 +δ
εδ 2 dϕ (cδ)2
Pseudoconvexity and Plurisubharmonicity
7
Lemma 1.2 Let Ω ⊂ C be an open set and let u be a continuous subharmonic function in Ω, a ∈ Ω. For r with 0 < r < dist(a, ∂Ω), define A(r) =
1 2π
2π
u(a + reiθ )dθ.
0
Then A(r) satisfies the following: 0 < r1 < r2 < dist(a, ∂Ω) =⇒ A(r1 ) ≤ A(r2 ). Proof. Let 0 < r < dist(a, ∂Ω). We denote by ϕr the Poisson integral of u. Then 2π 1 r2 − |z − a|2 ϕr (z) = u(a + reiθ )dθ. 2π 0 |(z − a) − reiθ |2 Moreover, ϕr is harmonic in B(a, r), continuous in B(a, r) and ϕr = u on ∂B(a, r). Then u(z) − ϕr (z) is subharmonic in B(a, r), and equals 0 on ∂B(a, r). By the maximum principle, u(z) ≤ ϕr (z) for z ∈ B(a, r). Therefore we have 2π 2π 1 1 u(a + r1 eiθ )dθ ≤ ϕr2 (a + r1 eiθ )dθ 2π 0 2π 0 2π 1 ϕr2 (a + r2 eiθ )dθ = 2π 0 2π 1 u(a + r2 eiθ )dθ. = 2π 0 Theorem 1.5 Let u : Ω → R be a continuous real-valued function in an open set Ω ⊂ C. Then the following statements are equivalent: (a) u is harmonic in Ω. (b) For any a ∈ Ω and any r with 0 < r < dist(a, ∂Ω), one has u(a) ≤
1 2π
2π
u(a + reiθ )dθ.
0
(c) For any a ∈ Ω, there exists ε (0 < ε < dist(a, ∂Ω)) such that for any r with 0 < r < ε one has 2π 1 u(a) ≤ u(a + reiθ )dθ. 2π 0
Several Complex Variables and Integral Formulas
8
(d) For any a ∈ Ω and any r with 0 < r < dist(a, ∂Ω), if h is continuous in |ζ − a| ≤ r, and harmonic in |ζ − a| < r, then u satisfies the following properties: u(ζ) ≤ h(ζ)
for
|ζ − a| = r =⇒ u(ζ) ≤ h(ζ)
for
|ζ − a| ≤ r.
Proof. (b)=⇒(c) and (a)=⇒(d) are trivial. We show that (d)=⇒(b). For a ∈ Ω, we choose r > 0 such that B(a, r) ⊂ Ω. We denote by U the Poisson integral of u for B(a, r). Then U is harmonic in B(a, r), continuous in B(a, r), and U (z) = u(z) for z ∈ ∂B(a, r). Since u ≤ U in B(a, r), we have 2π 2π 1 1 U (a + reiθ )dθ = u(a + reiθ )dθ. u(a) ≤ U (a) = 2π 0 2π 0 This proves (b). Next we show that (c)=⇒(a). Let K ⊂ Ω be a compact set. Suppose h is harmonic in the interior of K that is continuous on K, and satisfies u ≤ h on ∂K. We set c = max(u(z) − h(z)). z∈K
Suppose c > 0. We set Kc = {z ∈ K | u(z) − h(z) = c}. Then Kc is compact. We denote by a the nearest point of Kc to ∂K. If we choose r > 0 sufficiently small, then we have 1 2π
0
2π
1 {u(a + re ) − h(a + re )}dθ < 2π iθ
iθ
2π
cdθ = c.
0
On the other hand, if we choose r > 0 sufficiently small, then it follows from (c) that 2π 1 {u(a + reiθ ) − h(a + reiθ )}dθ, c = u(a) − h(a) ≤ 2π 0 which is a contradiction. Thus we have c = 0, which implies that u ≤ h on K. This proves (a). In order to prove the Hartogs theorem we need the following lemma (see Krantz [KR2]). Lemma 1.3 Let Ω ⊂ C be an open set and let f : Ω → R ∪ {−∞} be upper semicontinuous and bounded above. Then there exists a sequence
Pseudoconvexity and Plurisubharmonicity
9
{fj } of real-valued continuous functions in Ω which are bounded above on Ω such that f1 ≥ f2 ≥ · · · ,
fj → f.
Proof. In the case when f (x) ≡ −∞, we may set fn (x) = −n. Thus we may assume that f (x) ≡ −∞. For x ∈ Ω, we set fj (x) = sup{f (y) − j|x − y|}. y∈Ω
Then f1 (x) ≥ f2 (x) ≥ · · · ≥ f (x). For ε > 0, if x1 , x2 ∈ Ω, |x1 − x2 | < ε/j, then f (y) − j|x1 − y| < f (y) − j|x2 − y| + ε
(y ∈ Ω).
Therefore we have fj (x1 ) ≤ fj (x2 ) + ε. By interchanging x1 and x2 we have |fj (x1 ) − fj (x2 )| ≤ ε. Thus each fj is continuous in Ω. We set sup f (x) = M, x∈Ω
f (x) = α (α = −∞).
For ε > 0, there exists δ > 0 such that if |x − y| < δ, then f (y) < α − ε since f is upper semicontinuous. If |y − x| > δ, j > M/δ, then f (y) − j|x − y| ≤ M − M = 0. Thus we have M j> α = f (x) ≤ fj (x) = sup {f (y) − j|x − y|} < α + ε , δ |y−x|≤δ which shows that fj (x) → f (x). Suppose α = −∞. For N > 0, there exists δ1 > 0 such that f (y) < −N whenever |x − y| < δ1 . Hence we have
fj (x) = max
sup {f (y) − j|x − y|},
|x−y| N + M . Thus fj (x) → −∞ = f (x). Corollary 1.1 For an upper semicontinuous function u in an open set Ω ⊂ C, Theorem 1.5 also holds. Proof. We show that (d)=⇒(b). Let a ∈ Ω and 0 < r < r′ < dist(a, ∂Ω). By Lemma 1.3, there exists a sequence {uj } of continuous functions in B(a, r′ ) such that u1 ≥ u2 ≥ · · · ,
uj → u.
Several Complex Variables and Integral Formulas
10
We denote by Uj the Poisson integral of uj for B(a, r). Then Uj is harmonic in B(a, r), continuous in B(a, r), and satisfies Uj = uj on ∂B(a, r). Thus we have u(z) ≤ Uj (z) for z ∈ B(a, r). Therefore we obtain Z 2π Z 2π 1 1 Uj (a + reiθ )dθ = uj (a + reiθ )dθ. u(a) ≤ Uj (a) = 2π 0 2π 0 By letting j → ∞, (b) follows from the Fatou lemma. (c)=⇒(a) is proved in the same way. Definition 1.6 a ∈ Cn , define
The proof of
For rj > 0 (j = 1, · · · , n), we set r = (r1 , · · · , rn ). For
P (a, r) = {z = (z1 , · · · , zn ) | |zj − aj | < rj , j = 1, · · · , n}. If we set Pj = {zj ∈ C | |zj − aj | < rj }, then P (a, r) = P1 × · · · × Pn . P (a, r) is called a polydisc. When n = 1, we have P (a, r) = B(a, r). Definition 1.7 X ν
A power series of n variables is denoted by
cν (z − a)ν =
∞ X
cν1 ,··· ,νn (z1 − a1 )ν1 · · · (zn − an )νn .
(1.4)
ν1 =0,··· ,νn =0
The domain of convergence of the power series (1.4) is the interior of the set of points z ∈ Cn for which (1.4) converges. Theorem 1.6 Every power series converges uniformly on every compact subset of its domain of convergence. Proof. Let Ω be the domain of convergence of (1.4). For simplicity, we may assume that a = (a1 , · · · , an ) = 0. Let w ∈ Ω. Then we have sup |cν wν | = M < ∞. ν
We set r = (|w1 |, · · · , |wn |). Let K ⊂ P (0, r) be a compact set. Then there exists 0 < λ < 1 such that K ⊂ P (0, λr). If z ∈ P (0, λr), then we have |cν z ν | = |cν z1ν1 · · · znνn | ≤ |cν |(λr1 )ν1 · · · (λrn )νn = |cν wν |λ|ν| ≤ M λ|ν| .
Pseudoconvexity and Plurisubharmonicity
11
On the other hand we have
|ν|
λ
=
ν
1 1−λ
n
< ∞.
Thus ν cν z ν converges uniformly on K. Let E be any compact subset of Ω. Then there exists w1 , · · · , wk ∈ Ω such that for each wj , compact subsets Kj of polydiscs P (0, λrj ) constructed above satisfy k
E ⊂ ∪ Kj . j=1
Since (1.4) converges uniformly on each Kj , (1.4) converges uniformly on E. Definition 1.8 Let Ω ⊂ Cn be an open set. G ⊂⊂ Ω means that the closure of G in Cn is a compact subset of Ω. In this case, G is called relatively compact in Ω. Definition 1.9 Let Ω ⊂ Cn be an open set. A function f : Ω → C is called holomorphic in Ω if f is continuous in Ω, and for each a = (a1 , · · · , an ) ∈ Ω, if we set ϕ(zj ) = f (a1 , · · · , zj , · · · , an ), then ϕ(zj ) is holomorphic at aj . The set of all holomorphic functions in Ω is denoted by O(Ω). Theorem 1.7 For a function f : Ω → C on an open set Ω ⊂ Cn , the following statements are equivalent: (a) f is holomorphic in Ω. (b) Suppose f is continuous in Ω and P = P1 × · · · × Pn ⊂⊂ Ω. Then f (z) =
1 (2πi)n
∂P1
···
∂Pn
f (ζ)dζ1 · · · dζn (ζ1 − z1 ) · · · (ζn − zn )
(1.5)
for z ∈ P . (c) For any ξ ∈ Ω, there exists a neighborhood W of ξ such that f (z) =
∞
k1 =0,··· ,kn =0
for z ∈ W .
ak1 ,··· ,kn (z1 − ξ1 )k1 · · · (zn − ξn )kn
(1.6)
Several Complex Variables and Integral Formulas
12
(a)=⇒(b). By iterating the Cauchy integral formula, for z ∈ P f (ζ1 , z2 , · · · , zn ) 1 dζ1 f (z) = 2πi ∂P1 ζ1 − z1 1 1 f (ζ1 , ζ2 , z3 · · · , zn ) 1 = dζ2 dζ1 2πi ∂P1 ζ1 − z1 2πi ∂P2 ζ2 − z2 f (ζ1 , · · · , ζn ) 1 = dζ1 dζ2 · · · dζn . ··· (2πi)n ∂P1 ∂P2 (ζ − z1 ) · · · (ζn − zn ) 1 ∂Pn
Proof.
(b) =⇒ (c). Let ξ ∈ Ω. We choose r = (r1 , · · · , rn ) such that P = P (ξ, r) = P1 × · · · × Pn = {z | |zj − ξj | < rj (j = 1, · · · , n)} ⊂⊂ Ω. For ζ = (ζ1 , · · · , ζn ) ∈ ∂P1 × · · · × ∂Pn , since 1 1 = ζ1 − z1 (ζ1 − ξ1 )(1 −
z1 −ξ1 ζ1 −ξ1 )
=
∞ (z1 − ξ1 )k , (ζ1 − ξ1 )k+1 k=0
we obtain 1 = (ζ1 − z1 ) · · · (ζn − zn )
∞
k1 ,···kn =0
(z1 − ξ1 )k1 · · · (zn − ξn )kn . (ζ1 − ξ1 )k1 +1 · · · (ζn − ξn )kn +1
Since the power series of the right side of the above equality converges uniformly with respect to ζ, substituting into (1.5) and integrating, we obtain ∞ 1 f (ζ1 , · · · , ζn ) f (z) = n (2πi) ∂P1 ×···×∂Pn (ζ1 − ξ1 )k1 +1 · · · (ζn − ξn )kn +1 k1 ,··· ,kn =0
×dζ1 · · · dζn · (z1 − ξ1 )k1 · · · (zn − ξn )kn .
We set ak1 ,··· ,kn =
1 (2πi)n
∂P1 ×···×∂Pn
f (ζ1 , · · · , ζn ) dζ1 · · · dζn . (ζ1 − ξ1 )k1 +1 · · · (ζn − ξn )kn +1
Then f is expressed by f (z) =
∞
k1 ,··· ,kn =0
ak1 ,··· ,kn (z1 − ξ1 )k1 · · · (zn − ξn )kn .
This proves (c). (c)=⇒(a). We choose r > 0 such that {z | |zj − ξj | ≤ r, j = 1, · · · , n} ⊂ W.
Pseudoconvexity and Plurisubharmonicity
13
By Theorem 1.6, the right side of (1.6) converges uniformly on P (ξ, r). Therefore f is continuous in P (ξ, r). On the other hand, the finite sum N1
k1 =0
···
Nn
kn =0
ak1 ,··· ,kn (z1 − ξ1 )k1 · · · (zn − ξn )kn
is holomorphic in each variable zj and f is the uniform limit of the above finite sum when Nj → ∞. Thus f is holomorphic in P (ξ, r) with respect to each variable zj . Since ξ ∈ Ω is arbitrary, f is holomorphic in Ω. Definition 1.10 Let Ω ⊂ Cn be an open set and let f be holomorphic in Ω. For a multi-index α = (α1 , · · · , αn ), where each αj is a nonnegative integer, define |α| = α1 + · · · + αn , ∂ α f (z) =
α! = α1 ! · · · αn !,
∂ |α| f (z) · · · ∂znαn
∂z1α1
(z ∈ Ω).
Corollary 1.2 (Cauchy inequality) Let f be a holomorphic function in a polydisc P (0, r) = {z ∈ Cn | |zj | < rj , j = 1, · · · , n}. Suppose there exists a constant M > 0 such that |f (z)| ≤ M for z ∈ P (0, r). Then |∂ α f (0)| ≤ α!r−α M for any multi-index α = (α1 , · · · , αn ), where we define rα = r1α1 · · · rnαn . Proof. By Theorem 1.7 f is expressed by f (z) = follows from (1.6) that aα =
α
aα z α . Then it
∂ α f (0) . α!
On the other hand, by applying the proof of Theorem 1.7, for 0 < sj < rj , j = 1, · · · , n, we have f (ζ1 , · · · , ζn ) 1 dζ1 · · · dζn . aα = α n (2πi) {|z1 |=s1 }×···×{|zn |=sn } ζ1 1 +1 · · · ζn αn +1 Thus we obtain |∂ α f (0)| ≤ α!s−α M.
14
Several Complex Variables and Integral Formulas
The above inequality holds for any s > 0 which satisfies sj < rj for j = 1, · · · , n. Corollary 1.3 Let Ω ⊂ Cn be an open set and let K ⊂ Ω be a compact set. For any open subset ω of Ω with K ⊂ ω, there exists a constant Cα such that sup |∂ α f (z)| ≤ Cα uL1 (ω)
(f ∈ O(Ω)).
z∈K
Proof. In the case when ω is a polydisc, Corollary 1.3 follows from Theorem 1.2. In the general case, K is contained in the finite union of polydiscs which are contained in ω. Corollary 1.3 follows from Theorem 1.2. Lemma 1.4 (Baire’s theorem) Let X be a complete metric space. Then a countable intersection of open dense subsets of X is dense in X. Proof. Suppose {Vn } is a sequence of open dense subsets of X and W is an open nonempty subset of X. It is sufficient to show that ∩∞ n=1 Vn ∩ W is not empty. Let d(x, y) be the metric in X. We set B(x, r) = {y ∈ X | d(x, y) < r}. Since V1 ∩ W = φ, there exist x1 and r1 such that B(x1 , r1 ) ⊂ W ∩ V1 ,
0 < r1 < 1.
Since V2 ∩ B(x1 , r1 ) is not empty, there exist x2 and r2 such that B(x2 , r2 ) ⊂ V2 ∩ B(x1 , r1 ),
0 < r2
n. Then xi , xj ∈ B(xn , rn ), so that d(xi , xj ) < 2rn < 2/n, which implies that {xn } is a Cauchy sequence. Since X is complete, {xn } converges. Therefore, there exists x ∈ X such that x = limn→∞ xn . Since xi ∈ B(xn , rn ) (i > n) for each n, x ∈ B(xn , rn ) ⊂ Vn . Thus we have x ∈ ∩∞ n=1 Vn . On the other hand, x ∈ B(x1 , r1 ) ⊂ W , which shows that ∩∞ n=1 Vn is dense in X. Lemma 1.5 Suppose {vk } is a sequence of subharmonic functions in Ω. Assume that vk , k = 1, 2, · · · , are uniformly bounded from above on every compact subset of Ω and that there exists a constant C such that for any
Pseudoconvexity and Plurisubharmonicity
15
z ∈ Ω, limk→∞ vk (z) ≤ C. Then for any ε > 0 and any compact subset K of Ω there exists a positive integer N such that for k > N vk (z) ≤ C + ε
(z ∈ K).
Proof. We choose a compact set K1 with the properties that K ⊂ K1◦ ⊂ K1 ⊂ Ω. Since {vk } is uniformly bounded from above on K1◦ , we may assume that {vk } is uniformly bounded from above on Ω. Moreover, when vk (z) ≤ M for z ∈ Ω and M ≤ 0, we can treat vk − M instead of vk , so we may assume that vk (z) ≤ 0 for z ∈ Ω. We choose r > 0 such that K ⊂ {z ∈ Ω | dist(z, ∂Ω) > 3r}. By Corollary 1.1, for z ∈ K, 0 < ρ ≤ r, we have 2π 2πvk (z) ≤ vk (z + ρeiθ )dθ. (1.7) 0
If we multiply by ρ in (1.7) and integrate with respect to ρ from 0 to r, then we obtain vk (z ′ )dx′ dy ′ (z ′ = x′ + iy ′ ). (1.8) πr2 vk (z) ≤ |z−z ′ | 0 and z 0 such that maxj |zj0 − ζj | < R, ζn = zn0 , and f is bounded on P ′ = {z | |zj − zj0 | < r , j = 1, · · · , n − 1, |zn − zn0 | < R}(⊂ P (ζ, R)). Since f is holomorphic in P (z0 , R) by Lemma 1.9, f is holomorphic at ζ. 1.2
Characterizations of Pseudoconvexity
We prove that every domain of holomorphy is a pseudoconvex open set. In 2.2 (Corollary 2.4) we prove that an open set in Cn is a domain of holomorphy if and only if it is pseudoconvex. Definition 1.11
Let Ω ⊂ Cn be an open set.
(1) A real-valued upper semicontinuous function ϕ in Ω is called plurisubharmonic if for any v, w ∈ Cn , h(ζ) = ϕ(v + ζw) is subharmonic in U = {ζ ∈ C | v + ζw ∈ Ω}. The set of all plurisubharmonic functions in Ω is denoted by P S(Ω). (2) A real-valued C 2 function ϕ in Ω is called strictly plurisubharmonic in Ω if for any v, w ∈ Cn (w = 0), h(ζ) = ϕ(v+ζw) is strictly subharmonic in U = {ζ ∈ C | v + ζw ∈ Ω}. Theorem 1.9
Let Ω ⊂ Cn be an open set.
(a) If f ∈ O(Ω), then |f | ∈ P S(Ω).
Several Complex Variables and Integral Formulas
20
(b) If f is a holomorphic function in Ω and ρ is a C 2 subharmonic function in f (Ω), then ρ ◦ f ∈ P S(Ω). (c) Suppose {ρj }j∈J is a family of plurisubharmonic functions in Ω and ρ = supj∈J ρj . If ρ is finite and upper semicontinuos in Ω, then ρ ∈ P S(Ω). Proof. (a) For v, w ∈ C n , we set U = {ζ ∈ C | v + ζw ∈ Ω}. For ζ ∈ U , we set ϕ(ζ) = f (v + ζw). We fix a ∈ U . We choose r > 0 such that r < dist(a, ∂U ). Since ϕ is holomorphic in U , it follows from the Cauchy integral formula that ϕ(ζ) 1 dζ. ϕ(a) = 2πi |ζ−a|=r ζ − a Then we have 1 |ϕ(a)| ≤ 2π
2π 0
|ϕ(a + reiθ )|dθ.
Hence |ϕ| is subharmonic. Thus |f | is plurisubharmonic in Ω. (b) For v, w ∈ C n , we set U = {ζ ∈ C | v + ζw ∈ Ω}. For ζ ∈ U , we set ϕ(ζ) = ρ ◦ f (v + ζw). Then we have 2 ∂ ∂2ρ ∂2ϕ ≥ 0, (f (v + ζw)) (f (v + ζw)) (ζ) = ∂w∂ w ¯ ∂ζ ∂ζ∂ ζ¯
which means that ϕ is subharmonic in U . Thus ρ ◦ f ∈ P S(Ω). (c) For v, w ∈ C n , we set U = {ζ ∈ C | v + ζw ∈ Ω}. For ζ ∈ U , we set ϕj (ζ) = ρj (v + ζw) for j ∈ J. Since ϕj is subharmonic in U , it follows that 2π 1 ϕj (a + reiθ )dθ ϕj (a) ≤ 2π 0 for a ∈ U and r with 0 < r < dist(a, ∂U ). We set ϕ(ζ) = ρ(v + ζw). Then sup ϕj (ζ) = sup ρj (v + ζw) = ρ(v + ζw) = ϕ(ζ). j∈J
j∈J
For ε > 0, there exists j0 such that ϕ(a) − ε < ϕj0 (a). Therefore we obtain 2π 1 ϕ(a) < ϕj0 (a) + ε ≤ ϕj0 (a + reiθ )dθ + ε, 2π 0 1 ≤ 2π
0
2π
ϕ(a + reiθ )dθ + ε.
Pseudoconvexity and Plurisubharmonicity
21
Since ε > 0 is arbitrary, we obtain 2π 1 ϕ(a) ≤ ϕ(a + reiθ )dθ. 2π 0 Thus ϕ is subharmonic in U , which implies that ρ ∈ P S(Ω). Theorem 1.10
Let ρ be a real-valued C 2 function in an open set Ω ⊂ Cn .
(a) ρ is plurisubharmonic in Ω if and only if n
j,k=1
∂2ρ (z)wj w ¯k ≥ 0 ∂zj ∂ z¯k
for z ∈ Ω, w = (w1 , · · · , wn ) ∈ Cn . (b) ρ is strictly plurisubharmonic in Ω if and only if n
j,k=1
∂2ρ (z)wj w ¯k > 0 ∂zj ∂ z¯k
for z ∈ Ω, 0 = w = (w1 , · · · , wn ) ∈ Cn . Proof.
For v, w ∈ Cn , we set ρ˜(ζ) = ρ(v + ζw) = ρ(v1 + ζw1 , · · · , vn + ζwn ).
Then Theorem 1.10 follows from the equality n ∂ 2 ρ˜ ∂ 2 ρ˜ (v + ζw)wj w ¯k . (ζ) = ¯ ∂zj ∂ z¯k ∂ζ∂ ζ j,k=1
Corollary 1.4 Let Ω be an open set in Cn and let K be a compact subset of Ω. Suppose ρ is a strictly plurisubharmonic function in Ω. Then there exists a constant C = C(K) > 0 such that n
j,k=1
∂2ρ (z)wj w ¯k ≥ C|w|2 ∂zj ∂ z¯k
for z ∈ K, w = (w1 , · · · , wn ) ∈ Cn . Proof.
For z ∈ K, w ∈ Cn , we set f (z, w) =
n
j,k=1
∂2ρ (z)wj w ¯k . ∂zj ∂ z¯k
Several Complex Variables and Integral Formulas
22
Since f (z, w) takes the minimum value C > 0 on K × {w | |w| = 1|}, we have for z ∈ K and w = 0 n
j,k=1
Definition 1.12 z ∈ Cn , define
wj w ¯k ∂2ρ (z) ≥ C. ∂zj ∂ z¯k |w|2
Let Ω be an open subset of Cn and let Ω = Cn . For dist(z, ∂Ω) = inf{|z − ζ| | ζ ∈ ∂Ω},
where dist(z, ∂Ω) denotes the distance from z to ∂Ω. Lemma 1.10 Let Ω ⊂ Cn be an open set such that Ω = Cn . For z ∈ Ω, define ϕ(z) = dist(z, ∂Ω). Then ϕ is continuous in Ω. Proof. Fix z0 ∈ Ω. Then there exists ξ0 ∈ ∂Ω such that ϕ(z0 ) = |z0 −ξ0 |. For z ∈ Ω there exists ξ(z) ∈ ∂Ω such that ϕ(z) = |z − ξ(z)|. Then we have ϕ(z) = |z − ξ(z)| ≤ |z − ξ0 | ≤ |z − z0 | + |z0 − ξ0 | = |z − z0 | + ϕ(z0 ). Thus we have ϕ(z) − ϕ(z0 ) ≤ |z − z0 |. Similarly, we have ϕ(z0 ) = |z0 − ξ0 | ≤ |z0 − ξ(z)| ≤ |z0 − z| + ϕ(z). Hence we obtain |ϕ(z) − ϕ(z0 )| ≤ |z − z0 |, which means that ϕ is continuous at z = z0 .
Definition 1.13 Let Ω be an open subset in Cn . We say that Ω is a domain of holomorphy if there exists at least one holomorphic function in Ω that cannot be extended as a holomorphic function through any boundary point of Ω. Remark 1.1 nected.
We do not assume that the domain of holomorphy is con-
Pseudoconvexity and Plurisubharmonicity
23
Definition 1.14 Suppose Ω ⊂ Cn is an open set and K is a compact subset of Ω. Define O = {z ∈ Ω | |f (z)| ≤ sup |f (ζ)|, f ∈ O(Ω)}. K Ω ζ∈K
O is called a holomorphically convex hull of K (or O-hull of K). By K Ω O , K is called O(Ω)-convex. O . In case K = K definition, K ⊂ K Ω Ω
Definition 1.15 An open set Ω ⊂ Cn is called holomorphically convex if O ⊂⊂ Ω. (Equivalently, Ω is holomorphically for any compact subset K, K Ω O convex if and only if KΩ is compact for every compact subset K of Ω.) Lemma 1.11
Let Ω ⊂ Cn be an open set. Then
O ⊂ L O . (a) If K and L are compact subsets of Ω with K ⊂ L, then K Ω Ω O . If N is compact, then (b) We set N = K Ω
Proof.
O = N. N Ω
O . Then for any f ∈ O(Ω) we have (a) Let z ∈ K Ω |f (z)| ≤ sup |f (ζ)| ≤ sup |f (ζ)|. ζ∈K
ζ∈L
O . This proves (a). Hence f ∈ L Ω O , then O . If z ∈ N (b) By definition we have N ⊂ N Ω Ω |f (z)| ≤ sup |f (ζ)| ≤ sup |f (ζ)|. O ζ∈K Ω
ζ∈K
O ⊂ K O = N . This proves (b). Hence we have N Ω Ω
the smallest Lemma 1.12 Let K ⊂ Cn be compact. We denote by K is called the convex hull of K). Then we convex set which contains K (K O have KCn ⊂ K. Then there exists a hyperplane through w l : Proof. Suppose w ∈ K. 2n j=1 aj xj = b which does not intersect K. When zj = xj + ixn+j ∈ K, 2n 2n we assume that j=1 aj xj < b. If wj = uj + iun+j , then j=1 aj uj = b. If we set n αj = aj + ian+j , f (z) = exp α ¯ j zj − b , j=1
Several Complex Variables and Integral Formulas
24
then f ∈ O(Cn ). Using the equality Re
n
α ¯ j zj =
2n
aj xj ,
j=1
j=1
we have
|f (z)| = exp
2n j=1
aj xj − b < 1
|f (w)| = exp
Thus we have
2n j=1
(z ∈ K),
aj uj − b = 1.
sup |f (z)| < 1 = |f (w)|.
z∈K
On . Thus we obtain K On ⊂ K. Hence w ∈ K C C
Lemma 1.13 Suppose Ω ⊂ Cn is an open set and K is a compact subset O is bounded. of Ω. Then K Ω the convex hull of K. From Lemma 1.12 and Proof. We denote by K O ⊂ K On ⊂ K. the definition of holomorphically convex hull, we have K Ω C O Since K is bounded, KΩ is bounded. Definition 1.16 λ > 0, define
For r = (r1 , · · · , rn ), rj > 0, j = 1, · · · , n, a ∈ Cn and
P (a, λr) = {z | |zj − aj | < λrj j = 1, · · · , n}. For a ∈ Ω ⊂ Cn , we define (r)
δΩ (a) = sup{λ | λ > 0, P (a, λr) ⊂ Ω}. (r)
By definition, λ ≤ δΩ (a) if and only if P (a, λr) ⊂ Ω. Lemma 1.14
Let Ω = Cn be an open set. Then (r)
dist(a, ∂Ω) = inf{δΩ (a) | |r| = 1}
(a ∈ Ω).
Pseudoconvexity and Plurisubharmonicity
Proof.
25
We set (r)
δ = dist(a, ∂Ω),
η = inf{δΩ (a) | |r| = 1}.
If |r| = 1 and |zi − ai | < λri for i = 1, · · · , n, then |z − a| < λ, and hence P (a, λr) ⊂ B(a, λ). Thus we have P (a, δr) ⊂ Ω. Therefore, if |r| = 1, then (r) δ ≤ δΩ (a), which implies that δ ≤ η. Next we show that δ ≥ η. For any ε > 0, we choose λ such that δ < λ < δ + ε. Then B(a, λ) ⊂ Ω, which implies that there exists w such that w ∈ Ω, |w − a| < λ. We set |wi − ai | = ti ,
t = (t1 , · · · , tn ),
ri =
ti , |t|
r = (r1 , · · · , rn ).
Then |r| = 1, |wi − ai | < ri λ for i = 1, · · · , n. Hence w ∈ P (a, λr). (r) Therefore P (a, λr) ⊂ Ω, which means that δΩ (a) < λ. Thus we have (r) η ≤ δΩ (a) < λ < δ + ε. Since ε > 0 is arbitrary, we have η ≤ δ. Theorem 1.11 Suppose Ω ⊂ Cn is an open set and K is a subset of Ω. Let r = (r1 , · · · , rn ) and ri > 0 for i = 1, · · · , n. (r) that η > 0 satisfies δΩ (z) ≥ η for all z ∈ K. Then for any a ∈ f ∈ O(Ω), f is holomorphic in P (a, ηr). Proof.
compact Assume O and K Ω
We fix f ∈ O(Ω). We choose η ′ such that η ′ < η. We set Q = ∪ P (a, η ′ r). a∈K
Then Q is a compact subset of Ω. We set M = supz∈Q |f (z)|. By applying the Cauchy inequality for P (a, η ′ r), we have for α ∈ Nn sup |∂ α f (z)| ≤
z∈K
α!M . (η ′ r)α
O , we obtain Thus for a ∈ K Ω
|∂ α f (a)| ≤ sup |∂ α f (z)| ≤ z∈K
α!M . (η ′ r)α
Hence for z ∈ P (a, η ′ r), we have α ∂ α f (a) |z − a| α < ∞. M α! (z − a) ≤ η′ r α k
Thus
∂ α f (a) α α! (z
− a)α converges in P (a, η ′ r). If we set ϕ(z) =
∂ α f (a) α
α!
(z − a)α ,
Several Complex Variables and Integral Formulas
26
then ϕ is holomorphic in P (a, η ′ r). Since a ∈ Ω, we have ϕ = f in a neighborhood of a. Therefore f is holomorphic in P (a, η ′ r). Since η is arbitrary so far as η ′ < η, f is holomorhic in P (a, ηr). If Ω ⊂ Cn is a domain of holomorphy, then
Corollary 1.5
ΩO , ∂Ω) dist(K, ∂Ω) = dist(K
for every compact subset K of Ω.
O , we have dist(K, ∂Ω) ≥ dist(K O , ∂Ω). We set Proof. Since K ⊂ K Ω Ω O such that O , ∂Ω). We choose a ∈ K η = dist(K, ∂Ω). Let η > dist(K Ω Ω dist(a, ∂Ω) < η. It follows from Lemma 1.14 that there exists r such that (r) |r| = 1 and δΩ (a) < η. Therefore we have P (a, ηr) ⊂ Ω. On the other (r) hand, when |r| = 1 and z ∈ K, we have η ≤ dist(z, ∂Ω) ≤ δΩ (z). By Theorem 1.11, all f ∈ O(Ω) are holomorphic in P (a, ηr). Therefore f is holomorphic in a neighborhood of some boundary point of Ω. This contradicts that Ω is a domain of holomorphy. Next we state some properties of the infinite product which we will use in the proof of Theorem 1.12. Definition 1.17
Let {zn } be a sequence of complex numbers. We set Pn = (1 + z1 )(1 + z2 ) · · · (1 + zn ).
If limn→∞ Pn = P exist, then we define P =
∞
(1 + zn ).
(1.11)
n=1
The right side of (1.11) is called the infinite product. Lemma 1.15
Define PN =
N
(1 + zn ),
n=1
PN∗ =
N
Then (a) PN∗ ≤ exp(|z1 | + · · · + |zN |). (b) |PN − 1| ≤ PN∗ − 1. Proof.
(1 + |zn |).
n=1
Using the inequality 1 + |zi | ≤ e|zi | ,
Pseudoconvexity and Plurisubharmonicity
27
we have PN∗ =
N
(1 + |zn |) ≤ e|z1 |+···+|zN | .
n=1
This proves (a). We prove (b) by induction on N . When N = 1, it is trivial. Suppose (b) holds for N = k. Since Pk+1 − 1 = Pk (1 + zk+1 ) − 1 = (Pk − 1)(1 + zk+1 ) + zk+1 , we have |Pk+1 − 1| ≤ |Pk − 1| |1 + zk+1 | + |zk+1 |
≤ (Pk∗ − 1)(1 + |zk+1 |) + |zk+1 | ∗ = Pk+1 − 1.
Thus (b) holds for N = k + 1.
Lemma 1.16 Suppose {fk } is a sequence of bounded functions defined ∞ on a set E ⊂ Cn and j=1 |fj (z)| converges uniformly on E. Then
(a) Π∞ j=1 (1 + fj (z)) converges uniformly on E. (b) Let f = Π∞ j=1 (1 + fj ) and z0 ∈ E. Then f (z0 ) = 0 if and only if there exists n such that fn (z0 ) = −1. (c) Let {k1 , k2 , · · · } be a permutation of {1, 2, · · · }. Then ∞
(1 + fj ) =
j=1
∞
(1 + fkj ).
j=1
Proof. Since fk is bounded on E, there exists a constant ck such that |fk (z)| ≤ ck for z ∈ E. We set h(z) =
∞ j=1
|fj (z)|,
hm (z) =
m j=1
|fj (z)|.
Since {hm } converges to h uniformly on E, there exists a positive integer n0 such that for m ≥ n0 |h(z) − hm (z)| < 1
(z ∈ E).
Several Complex Variables and Integral Formulas
28
Hence for z ∈ E, we have |h(z)| ≤ |hn0 (z)| + 1 =
n0 j=1
|fj (z)| + 1 ≤
n0
cj + 1 =: c˜.
j=1
By Lemma 1.15, if we set Pm (z) =
m
∗ Pm (z) =
(1 + fj (z)),
j=1
m
j=1
(1 + |fj (z)|),
then we have ∗ (z)| ≤ exp(|f1 (z)| + · · · + |fm (z)|) = ehm (z) ≤ eh(z) ≤ ec˜ =: c. |Pm (z)| ≤ |Pm
Let 0 < ε < 1/2. Then there exists a positive integer t0 such that |ht0 (z) − h(z)| =
∞
k=t0 +1
|fk (z)| < ε.
Let N ≥ t0 . Since {k1 , k2 , · · · } is a permutation of {1, 2, · · · }, we have for some sufficiently large integer M {1, 2, · · · , N } ⊂ {k1 , k2 , · · · , kM }. We set qM (z) =
M
(1 + fkj (z)).
j=1
We set F = {k1 , k2 , · · · , kM } − {1, 2, · · · , N }. Then we have qM (z) − PN (z) =
M
j=1
(1 + fkj (z)) −
= PN (z)
i∈F
N
j=1
(1 + fj (z))
(1 + fi (z)) − 1 .
By Lemma 1.15 we obtain |fi (z)| − 1. (1 + |fi (z)|) − 1 ≤ exp (1 + fi (z)) − 1 ≤ i∈F
i∈F
i∈F
Pseudoconvexity and Plurisubharmonicity
For z ∈ E, we have
|qM (z) − PN (z)| ≤ |PN (z)| exp
∞
i=t0 +1
29
|fi (z)|
−1
≤ |PN (z)|(eε − 1) ε2 + ···) = |PN (z)|(ε + 2! 1 1 ≤ ε|PN (z)|(1 + + 2 + · · · ) 2 2 = 2ε|PN (z)| ≤ 2εc. Thus for N with N ≥ t0 and some sufficiently large M we have |qM (z) − PN (z)| ≤ 2ε|PN (z)| ≤ 2εc
(z ∈ E).
(1.12)
In particular, when kj = j we have qM = PM . It follows from (1.12) that |PM (z) − PN (z)| < 2εc
(z ∈ E).
Thus {PN } converges uniformly on E. This proves (a). Let kj = j. Then for some sufficiently large M it follows from (1.12) that |PM (z) − Pt0 (z)| ≤ 2ε|Pt0 (z)|. Thus we obtain |Pt0 (z)|(1 − 2ε) ≤ |PM (z)|. Letting M → ∞ we have |Pt0 (z)|(1 − 2ε) ≤ |f (z)|. Since 1 − 2ε > 0, f (z0 ) = 0 implies Pt0 (z0 ) = 0. Thus there exists k such that fk (z0 ) = −1. This proves (c). From (a), {Pj (z)} converges to f uniformly on E. Therefore, taking N sufficiently large with N ≥ t0 , if necessary, we have |f (z) − PN (z)| < ε
(z ∈ E).
For some sufficiently large M , we have |qM (z) − f (z)| ≤ |qM (z) − PN (z)| + |PN (z) − f (z)| < 2εc + ε = ε(1 + 2c). Thus we have limM→∞ qM (z) = f (z).
Several Complex Variables and Integral Formulas
30
Theorem 1.12 equivalent: (a) (b) (c) (d)
For an open set Ω ⊂ Cn , the following statements are
Ω is a domain of holomorphy. O is compact. For any compact set K ⊂ Ω, K Ω O , ∂Ω). For any compact set K ⊂ Ω, dist(K, ∂Ω) = dist(K Ω If X ⊂ Ω is a discrete infinite subset of Ω, then there exists f ∈ O(Ω) such that f is unbounded on X.
Proof. (a) =⇒ (c) follows from Theorem 1.11. (c)=⇒(b). Since K is compact, we have O , ∂Ω) ≥ dist(K, ∂Ω) > 0. dist(K Ω
O is compact. O is closed in Ω, K Since K Ω Ω (b)=⇒(a). Let {Kn } be a sequence of compact sets such that Ω = ◦ ◦ ∪∞ n=1 Kn with Kn ⊂ Kn+1 , where Kn denotes the interior of Kn . It O n+1 )O . If we set Tn = (K n )O , follows from Lemma 1.11 that (Kn )Ω ⊂ (K Ω Ω ∞ then by the assumption, Tn is compact and Ω = ∪n=1 Tn . It follows from Lemma 1.11 that Tn ⊂ Tn+1 , (Tn )O Ω = Tn . We may assume that Tn ⊂ ◦ Tn+1 . Suppose X ⊂ Ω is a countable set and X = Ω. Let X = {ξm }∞ m=1 . We denote by Bm the largest open ball with center ξm and contained in Ω. Let ηm ∈ Bm − Tm . Since η ∈ Tm , there exists fm ∈ D such that |fm (ηm )| > sup |fm (ζ)|. ζ∈Tm
We set gm (z) =
fm (z) . fm (ηm )
Then gm ∈ O(Ω) and gm (ηm ) = 1, supζ∈Tm |gm (ζ)| < 1. For some sufficiently large integer km km (ζ)| < sup |gm
ζ∈Tm
1 , m2m
km (ηm ) = 1. gm
km . Then ϕm ∈ O(Ω), ϕm (ηm ) = 1 and supζ∈Tm |ϕm (ζ)| < Set ϕm = gm m −1 (m2 ) . We set
ϕ(z) =
∞
j=1
(1 − ϕj (z))j = (1 − ϕ1 (z))(1 − ϕ2 (z))(1 − ϕ2 (z))(1 − ϕ3 (z)) · · · .
Pseudoconvexity and Plurisubharmonicity
Then for z ∈ Tm m|ϕm (z)| + (m + 1)|ϕm+1 (z)| + · · · ≤
31
1 1 + m+1 + · · · . 2m 2
∞ Therefore, for any positive integer m, j=1 j|ϕj (z)| converges uniformly on ∞ Tm . Thus, j=1 (1 − ϕj (z))j converges uniformly on every Tm . Thus, ϕ is holomorphic in Ω. Since |ϕm (z)| < 1 for z ∈ T1 , ϕ(z) = 0 for z ∈ T1 . Thus ϕ(z) ≡ 0. Suppose that there exists a domain V such that φ = Ω ∩ V = V and that ϕ is holomorphic in Ω∪V . We set V ∩Ω = W . Let ζ ∈ ∂W ∩∂Ω∩V . Since X ∩ W is dense in W , We can choose a subsequence {ξmj } of X which converges to ζ. If we choose j sufficiently large, then Bmj ⊂ W . Since ηmj ∈ Bmj − Tmj , we have ηmj → ζ. In case k = (k1 , · · · , kn ) with |k| = k1 + · · · + kn < mj , we have |k| |k| ∂ ϕ ∂ mj m (z)) . (1 − ϕ (z)) (1 − ϕ (z) = mj m ∂z k ∂z k m=mj
Hence we have
∂kϕ ∂z1k1 · · · ∂znkn
(ηmj ) = 0
(k < mj , k = k1 + · · · + kn ).
Since ζ ∈ V and ϕ is holomorphic in V , we obtain 0=
∂kϕ ∂z1k1 · · · ∂znkn
(ζ).
Thus ϕ(z) ≡ 0 in Ω ∪ V . This is a contradiction. (d)=⇒(b). Suppose (b) is not true. There exists a compact set K ⊂ Ω O is a closed subset of Ω with respect O is not compact. Since K such that K Ω Ω O , k = 1, 2, · · · , such that {ξk } to the relative topology, there exist ξk ∈ K Ω converges to a boundary point of Ω. Then we have for any f ∈ O(Ω) |f (ξk )| ≤ sup |f (z)| < ∞. z∈K
Since X = {ξi } is a discrete infinite subset in Ω and f is bounded in X, (d) does not hold. (b)=⇒(d). We choose a sequence {Kn } of compact subsets of Ω such m )O are that Ω = ∪∞ Km , Km ⊂ Km+1 . By the assumption , Tm = (K m=1
Ω
compact and satisfy Ω = ∪∞ m=1 Tm , Tm ⊂ Tm+1 . We may assume that Tm ⊂ (Tm+1 )◦ . Suppose X ⊂ Ω is a discrete infinite set. Let X = {ξm }. We choose a subsequence {Tmj } of {Tn } and a subsequence {ξνj } of {ξm }
Several Complex Variables and Integral Formulas
32
such that ξνj ∈ Tmj+1 − Tmj . For simplicity, we rewrite ξνj by ξj and Tmj by Tj . Hence ξj ∈ Tj+1 − Tj . Since (Tj )O Ω = Tj ∋ ξj , there exist fj ∈ O(Ω) such that |fj (ξj )| > sup |fj (ζ)|. ζ∈Tj
Choose αj such that |fj (ξj )| > αj > sup |fj (ζ)|. ζ∈Tj
We set hj = fj /αj . Then we have |hj (ξj )| > 1, supζ∈Tj |hj (ζ)| < 1. For k
any sufficiently large integer kj , We set ϕj = hj j . Then we have ϕj ∈ O(Ω) and sup |ϕj (ζ)|
j + 1 +
∞
j−1
k=1
|ϕk (ξj )|
(j = 1, 2, · · · ).
ϕk , then ϕ ∈ O(Ω). Hence we obtain ∞ |ϕ(ξj )| = ϕk (ξj ) ≥ |ϕj (ξj )| − ϕk (ξj ) k=1
k=1
≥ j+1+ = j+1−
k=j
j−1
k=1
k>j
|ϕk (ξj )| −
k=j
|ϕk (ξk )|
|ϕk (ξj )|.
Since ξj ∈ Tj+1 , we have ξj ∈ Tk for k ≥ j + 1. Thus, we have |ϕk (ξj )| ≤ sup |ϕk (ζ)| < ζ∈Tk
1 , 2k
for k ≥ j + 1, which means that 1 < 1. |ϕk (ξj )| ≤ 2k k>j
k>j
Since |ϕ(ξj )| ≥ j, we have limj→∞ |ϕ(ξj )| = ∞. Thus ϕ is unbounded on X. Definition 1.18 (1) Let Ω ⊂ Cn be an open set such that Ω = Cn . Ω is called a pseudoconvex open set if − log dist (z, ∂Ω) is plurisubharmonic in Ω. In particular, we define Cn to be pseudoconvex.
Pseudoconvexity and Plurisubharmonicity
33
(2) Let Ω be a bounded open set in Cn . Ω is called strictly pseudoconvex if there exist a neighborhood W of ∂Ω and a strictly plurisubharmonic function ρ in W such that Ω ∩ W = {z ∈ W | ρ(z) < 0}. Lemma 1.17 (a) Let Ω ⊂ Cn and G ⊂ Cm . Suppose Ω and G are pseudoconvex open sets. Then Ω × G is a pseudoconvex open set in Cn+m . (b) Let {Ωj }j∈J be a family of pseudoconvex open sets in Cn . Then the interior (∩j∈J Ωj )◦ of ∩j∈J Ωj is a pseudoconvex open set. Proof.
(a) We have ∂(Ω × G) = (∂Ω × G) ∪ (Ω × ∂G).
Hence for (z, w) ∈ Ω × G, we have dist((z, w), ∂(Ω × G)) = min{dist(z, ∂Ω), dist(w, ∂G)}. Consequently, − log dist((z, w), ∂(Ω × G)) = − inf{log dist(z, ∂Ω), log dist(w, ∂G)}. Then − log dist((z, w), ∂(Ω×G)) is plurisubharmonic in Ω×G, which implies that Ω × G is pseudoconvex. ◦ (b) We set Ω = (∩j∈J Ωj ) . For z ∈ Ω, we have dist(z, ∂Ω) = inf j∈J dist(z, ∂Ωj ). Hence we obtain − log dist(z, ∂Ω) = sup{− log dist(z, ∂Ωj )}. j∈J
Then − log dist(z, ∂Ω) is plurisubharmonic in Ω. Definition 1.19 Define
Suppose Ω ⊂ Cn is an open set and K ⊂ Ω is compact.
ΩP = {z ∈ Ω | ρ(z) ≤ max ρ(ζ), ρ ∈ P S(Ω)}. K ζ∈K
P . In case K = K P, P is a closed subset in Ω and K ⊂ K By definition, K Ω Ω Ω we say that K is P S(Ω)-convex. Lemma 1.18 Then
Suppose Ω ⊂ Cn is an open set and K ⊂ Ω is compact. ΩP ⊂ K ΩO . K
Proof. By Theorem 1.9, if f ∈ O(Ω), then |f | ∈ P S(Ω), which completes the proof of Lemma 1.18.
Several Complex Variables and Integral Formulas
34
Theorem 1.13 Suppose Ω is an open set in Cn . Then the following statements are equivalent: (a) Ω is pseudoconvex. b P is compact. (b) If K ⊂ Ω is compact, then K Ω (c) There exists ρ ∈ P S(Ω) such that for any real number α, the closure in Ω of the set Ωα := {z ∈ Ω | ρ(z) < α} is compact. Proof. (a) In case Ω 6= Cn . (a)=⇒(c). We set ρ(z) = max{|z|2 , − log dist(z, ∂Ω)}. Then ρ ∈ P S(Ω). Let ρ(z) < α. Then we have |z|2 < α,
− log dist(z, ∂Ω) < α.
Thus we have dist(z, ∂Ω) > e−α , which means that Ωα ⊂⊂ Ω. (c)=⇒(b). Suppose there exists ρ1 ∈ P S(Ω) such that {z ∈ Ω | ρ1 (z) < α} ⊂⊂ Ω for any real α. Let K ⊂ Ω be compact. If we choose α such that α = supζ∈K ρ1 (ζ) + 1, then b ΩP ⊂ {z ∈ Ω | ρ1 (z) ≤ sup ρ1 (ζ)} ⊂⊂ Ω, K ζ∈K
b P is a compact subset of Ω. which implies that K Ω (b)=⇒(a). We set ϕ(z) = − log dist(z, ∂Ω). We show that ϕ(z) is plurisubharmonic in Ω. For v, w ∈ Cn and w 6= 0, we set U = {λ ∈ Cn | v + λw ∈ Ω}. We set g(λ) = ϕ(v + λw). It is sufficient to show that g(λ) is subharmonic in U . For λ0 ∈ U , it is sufficient to show that g(λ) is subharmonic in a neighborhood of λ0 . We set a = v + λ0 w. Then a ∈ Ω. Since a + λw = v + w(λ + λ0 ), if we set ψ(λ) = ϕ(a + λw), then ψ(λ) = g(λ + λ0 ). So it is sufficient to show that ψ(λ) is subharmonic in a neighborhood of 0. There exists r > 0 such that {a + λw | |λ| ≤ r} ⊂ Ω. Let h be harmonic in a neighborhood of |λ| ≤ r. It is sufficient to show that if h(λ) ≥ ψ(λ) on |λ| = r, then h(λ) ≥ ψ(λ) on |λ| ≤ r. There exists a harmonic function h∗
Pseudoconvexity and Plurisubharmonicity
35
in |λ| ≤ r such that f = h + ih∗ is holomorphic in |λ| ≤ r. We have for |λ| = r eh(λ) ≥ eψ(λ) = e− log dist(a+λw,∂Ω) =
1 . dist(a + λw, ∂Ω)
Thus on |λ| = r we have dist(a + λw, ∂Ω) ≥ e−h(λ) = |e−f (λ) |. Therefore, if |λ| = r, |ζ| < 1, then a + λw + ζe−f (λ) ∈ Ω. We fix ζ with |ζ| < 1. For 0 ≤ t ≤ 1, we set Γt = {a + λw + tζe−f (λ) | |λ| ≤ r}. Then we have Γ0 = {a + λw | |λ| ≤ r} ⊂ Ω. We set T = {t ∈ [0, 1] | Γt ⊂ Ω}. Then T ⊂ [0, 1], 0 ∈ T . If T is closed and open in [0, 1], then T = [0, 1], and hence 1 ∈ T . Then for |ζ| < 1 we have Γ1 = {a + λw + ζe−f (λ) | |λ| ≤ r} ⊂ Ω. Thus, for |λ| ≤ r, we obtain dist(a + λw, ∂Ω) ≥ |e−f (λ) | = e−h(λ) , which implies that ψ(λ) ≤ h(λ)
(|λ| ≤ r).
Hence ψ(λ) is subharmonic. Finally, we show that T is closed and open in [0, 1]. Let t0 ∈ T . Then Γt0 ⊂ Ω. Since Ω is open, Γt ⊂ Ω for any sufficiently closed point t to t0 . Thus t ∈ T , and hence T is open. Next we show that T is closed. We set K = {a + λw + tζe−f (λ) | |λ| = r, 0 ≤ t ≤ 1}. b P is compact. Let Then K is compact and K ⊂ Ω. By the assumption, K Ω −f (λ) t ∈ T . We set g(λ) = a + λw + tζe . Since Γt ⊂ Ω, g(λ) ∈ Ω for |λ| ≤ r. Hence g(λ) is holomorphic in |λ| ≤ r. Let ρ ∈ P S(Ω). Then
Several Complex Variables and Integral Formulas
36
ρ ◦ g(λ) is subharmonic in |λ| ≤ r. By applying the maximum principle for subharmonic functions, we have for |λ| ≤ r ρ ◦ g(λ) ≤ sup ρ ◦ g(λ) = sup ρ(a + λw + ζe−f (λ) ) ≤ sup ρ(z). |λ|=r
z∈K
|λ|=r
P . Hence for t ∈ T we obtain Thus we have g(λ) ∈ K Ω P. Γt = {g(λ) | |λ| ≤ r} ⊂ K Ω
P . Since K P is compact, we have Let tν ∈ T and tν → t0 . Then Γtν ⊂ K Ω Ω P Γt0 ⊂ KΩ ⊂ Ω, and hence t0 ∈ T . Thus T is closed. Hence T is closed and open in [0, 1], which shows that (b)=⇒(a). (b) In case Ω = Cn . By definition, Ω is pseudoconvex. If K ⊂ Cn is P ⊂ K O, K P is O is bounded. Since K compact, then by Lemma 1.13 K Ω Ω Ω Ω 2 n compact. We set ρ(z) = |z| . Then ρ ∈ P S(Ω) and Ωα = {z ∈ C | ρ(z) < α} ⊂⊂ Cn . Corollary 1.6 Let Ω ⊂ Cn be an open set. If Ω is a domain of holomorphy, then Ω is pseudoconvex. O Proof. Let K ⊂ Ω be compact. It follows from Theorem 1.12 that K Ω P O P is compact. Since KΩ ⊂ KΩ , KΩ is compact. By Theorem 1.13, Ω is pseudoconvex.
Lemma 1.19 (Dini’s theorem) Suppose K is a compact subset in Cn and that {fn } is a sequence of real-valued continuous functions on K that converges to f monotonically on K. Then {fn } converges to f uniformly on K. Proof.
Suppose f1 (x) ≥ f2 (x) ≥ · · · ,
fn (x) → f (x).
We set gn (x) = fn (x) − f (x). We denote by αn the maximum of gn in K. Then αn is monotonically decreasing. Let limn→∞ αn = α. It is sufficient to show that α = 0. Suppose α > 0. Let xn ∈ K be a point such that αn = gn (xn ). We can choose a convergent subsequence {xkn } of {xn }. Define limn→∞ xkn = x0 . Then we have gkn (x0 ) → 0 as n → ∞. If we choose N sufficiently large, then we obtain n ≥ N =⇒
gkn (x0 )
1, then λ(z) = 0. (2) λ ! depends only on |z1 |, · · · , |zn |. (3) Cn λ(z)dV (z) = 1, where dV denotes the Lebesgue measure on Cn .
Let Ω be an open set in Cn and let u be a plurisubharmonic function in Ω. For ε > 0, we set Ωε = {z ∈ Ω | dist(z, ∂Ω) > ε}
and uε (z) =
|ζ| 0, χ′′ (t) > 0 (for example, χ(t) = te− t (t > 0), 0 (t ≤ 0)). Define Ψj = χ(Φj + 2 − j).
Pseudoconvexity and Plurisubharmonicity
39
Then Ψj is strictly plurisubharmonic in a neighborhood of Ωj \Ωj−1 and Ψj > 0. Φ0 is strictly plurisubharmonic and Φ0 ≥ Φ in a neighborhood of Ω0 . Since Ψ1 is strictly plurisubharmonic and Ψ1 > 0 in a neighborhood of Ω1 \Ω0 , Φ0 + a1 Ψ1 > Φ in a neighborhood of Ω1 \Ω0 if a1 > 0. Further, by Corollary 1.4 there exists a constant C > 0 such that n ∂ 2 Ψ1 (z)wi w ¯j ≥ C|w|2 . ∂z ∂ z ¯ i j i,j=1
Similarly, there exists a constant C1 > 0 such that n ∂ 2 Φ0 (z)wi w ¯j ≤ C1 |w|2 . i,j=1 ∂zi ∂ z¯j Hence for any sufficiently large a1 > 0, we have
n n ∂ 2 Ψ1 ∂ 2 Φ0 (z)wi w ¯ j + a1 (z)wi w ¯j ≥ a1 C|w|2 − C1 |w|2 > 0. ∂z ∂ z ¯ ∂z ∂ z ¯ i j i j i,j=1 i,j=1
Hence u1 = Φ0 + a1 Ψ1 is strictly plurisubharmonic in a neighborhood of Ω1 \Ω0 . Since Φ0 is strictly plurisubharmonic, Φ0 ≥ Φ in a neighborhood of Ω0 and Ψ1 ≥ 0 in a neighborhood of Ω1 , u1 is strictly plurisubharmonic and u1 > Φ in a neighborhood of Ω1 . Repeating this process, there exist positive numbers a1 , · · · , am such that u m = Φ0 +
m
aj Ψ j
j=1
is strictly plurisubharmonic and um > Φ in a neighborhood of Ωm . If k ≥ j + 3, then Ψk = 0 on Ωj . Thus there exists u = limm→∞ um such that u is strictly plurisubharmonic, u ∈ C ∞ (Ω) and u ≥ Φ in Ω. Lemma 1.20 Let f be differentiable at x = a and let f (a) = 0. Let h be continuous at x = a. Then f h is differentiable at x = a. Moreover, we have {f (x)h(x)}′x=a = h(a)f ′ (a). Proof. lim
x→a
By the definition of differentiation, we have h(x)(f (x) − f (a)) h(x)f (x) − h(a)f (a) = lim = h(a)f ′ (a). x→a x−a x−a
Several Complex Variables and Integral Formulas
40
Definition 1.20 Let Ω ⊂ Rn be an open set. We say that Ω has a C k (k ≥ 1) boundary if there exist a neighborhood U of ∂Ω and a C k function ρ in U such that (1) Ω ∩ U = {x ∈ U | ρ(x) < 0}. (2) dρ = 0 on ∂Ω, where dρ(x) =
n ∂ρ (x)dxj . ∂x j j=1
Lemma 1.21 Let Ω = {x | ρ(x) < 0} ⊂ Rn be a bounded domain with C k (k ≥ 1) boundary and let f be a C k function in a neighborhood of Ω. Assume that f (x) = 0 for all x ∈ ∂Ω. Then for P ∈ ∂Ω there exist a neighborhood U of P and a C k−1 function h in U such that f (x) = ρ(x)h(x) for x ∈ U . Proof. Without loss of generality, we may assume that P = 0. Since dρ = 0 on ∂Ω, we may assume that there exists a neighborhood U of P such that if x = (x′ , xn ) (x′ = (x1 , · · · , xn−1 )) forms a coordinate system in U . Then we have ρ(x) = xn for x ∈ U . Since f (x′ , 0) = 0, we have 1 d f (x′ , xn ) = f (x′ , xn ) − f (x′ , 0) = {f (x′ , txn )}dt dt 0 1 ∂f ′ (x , txn )dt. = xn 0 ∂xn Define ′
h(x , xn ) = ′
Then h(x , xn ) is of class C
k−1 n
1 0
∂f ′ (x , txn )dt. ∂xn
in U and f (x) = ρ(x)h(x) for x ∈ U .
2
Theorem 1.16 Let Ω ⊂ C be an open set with C boundary. Let Ω = of Ω and | ρ(z) < 0}, where ρ is a C 2 function in a neighborhood Ω {z ∈ Ω satisfies dρ = 0 on ∂Ω. Then Ω is pseudoconvex if and only if n
j,k=1
∂2ρ (z)wj w ¯k ≥ 0 ∂zj ∂ z¯k
for all z and w = (w1 , · · · , wn ) satisfying z ∈ ∂Ω,
n ∂ρ (z)wj = 0. ∂z j j=1
(1.13)
Pseudoconvexity and Plurisubharmonicity
41
Proof. Suppose ρ1 is a C 2 defining function for Ω. By Lemma 1.21 there exists a C 1 function h in a neighborhood V of ∂Ω such that ρ1 = hρ. Since dρ1 = hdρ on ∂Ω, we have h > 0 in V . For z and w satisfying z ∈ ∂Ω and n ∂ρ1 j=1 ∂zj (z)wj = 0, we have n ∂ρ (z)wj = 0. ∂z j j=1
By Lemma 1.20, we obtain n
j,k=1
n ∂2ρ ∂ 2 ρ1 (z)wj w ¯k = h(z) (z)wj w ¯k ≥ 0. ∂zj ∂ z¯k ∂zj ∂ z¯k j,k=1
Thus the condition (1.13) is independent of the choice of the defining function ρ. Suppose Ω is pseudoconvex. Define −dist(z, ∂Ω) (z ∈ Ω) ρ˜(z) = dist(z, ∂Ω) (z ∈ Ωc ). Then ρ˜ is a C 2 function in a neighborhood of ∂Ω and satisfies d˜ ρ = 0 on ∂Ω (see Krantz-Parks [KRP]). If z ∈ Ω is sufficiently close to ∂Ω, then for δ(z) := dist(z, ∂Ω), − log δ(z) is plurisubharmonic. By Theorem 1.10, we have n
j,k=1
∂˜2 (− log δ(z))wj w ¯k ∂zj ∂ z¯k
2 n n 2 ∂δ ∂ δ 1 1 (z)wj w ¯k + (z)w =− j ≥ 0. 2 δ ∂zj ∂ z¯k δ(z) j=1 ∂zj j,k=1
Thus if z ∈ Ω is sufficiently close to ∂Ω, then 2 n n ∂ 2 ρ˜ 1 ∂δ (z)wj w ¯k + (z)wj ≥ 0. ∂zj ∂ z¯k δ(z) j=1 ∂zj j,k=1
Suppose that z ∈ ∂Ω, (i)
{w } satisfying z
(i)
∈ Ω,
z
n
(i)
∂ ρ˜ j=1 ∂zj (z)wj
→ z,
w
(i)
(1.14)
= 0. We choose sequences {z (i) } and
→ w,
n ∂ ρ˜ (i) (i) (z )wj = 0. ∂z j j=1
42
Several Complex Variables and Integral Formulas
By (1.14) we have n X
j,k=1
∂ 2 ρ˜ (i) (i) ¯k ≥ 0. (z (i) )wj w ∂zj ∂ z¯k
Letting i → ∞ we have (1.13). Conversely we assume that (1.13) holds. We set ϕ(τ ) = log δ(z + τ w). It is sufficient to show that −ϕ(τ ) is subharmonic. Assume that ∂2ϕ (0) = c > 0. ∂τ ∂ τ¯ Using Taylor’s formula, we obtain 2 ∂2ϕ ∂ϕ ∂ ϕ 2 + (0)τ ϕ(τ ) = ϕ(0) + 2Re (0)τ + Re (0)|τ |2 + o(|τ |2 ). 2 ∂τ ∂τ ∂τ ∂ τ¯ We set A=2
∂ϕ (0), ∂τ
B=
∂2ϕ (0). ∂τ 2
Then ϕ(τ ) = log δ(z) + Re(Aτ + Bτ 2 ) + c|τ |2 + o(|τ |2 ). Suppose z0 ∈ ∂Ω satisfies δ(z) = |z − z0 |. For 0 < s ≤ 1, define 2
ψs (τ ) = z + τ w + s(z0 − z)eAτ +Bτ . Then 2
δ(ψs (τ )) = δ(z + τ w + s(z0 − z)eAτ +Bτ ) 2
≥ δ(z + τ w) − s|z − z0 ||eAτ +Bτ |. On the other hand, we have 2
δ(z + τ w) = eϕ(τ ) = δ(z)|eAτ +Bτ |ec|τ |
2
+o(|τ |2)
,
which implies that 2
δ(ψs (τ )) ≥ δ(z)|eAτ +Bτ |ec|τ | 2
2
/2
= δ(z)|eAτ +Bτ |(ec|τ |
2
2
− sδ(z)|eAτ +Bτ | /2
− s).
For s with 0 < s < 1 we have ψs (0) = z +s(z0 −z) ∈ Ω. Hence for 0 < s < 1 and any sufficiently small |τ |, we have ψs (τ ) ∈ Ω, and hence ψ1 (τ ) ∈ Ω. If
Pseudoconvexity and Plurisubharmonicity
43
we set f (τ ) = δ(ψ1 (τ )), then f (0) = 0. So we have for any sufficiently small |τ | f (τ ) ≥
2 c δ(z)|eAτ +Bτ ||τ |2 . 4
Thus f (τ ) takes a local minimum at τ = 0, and hence by (1.15), the case that formula, we have f (τ ) = Re
2
∂ f ∂τ 2 (0)
=
2
∂ f ∂τ ∂ τ¯ (0)
∂2f (0)τ 2 ∂τ 2
+
(1.15) ∂f ∂τ (0)
= 0. Further,
= 0 does not occur. By Taylor’s
∂2f (0)|τ |2 + o(|τ |2 ). ∂τ ∂ τ¯
In the above equation we set τ = eit λ, where t and λ are real numbers. ∂2f Then in case ∂τ ∂ τ¯ (0) ≤ 0, f (τ ) is negative for some t, which implies that ∂2 f ∂τ ∂ τ¯ (0)
> 0. For any sufficiently small |τ |, we have ψ1 (τ ) ∈ Ω, and hence ρ˜(ψ(τ )) = −δ(ψ(τ )) = −f (τ ). Thus if we set ψ1 (τ ) = λ(τ ), then n X ∂ ρ˜ (z0 )λ′j (0) = 0 ∂z j j=1
and n X
j,k=1
∂ 2 ρ˜ (z0 )λ′j (0)λ′j (0) < 0. ∂zj ∂ z¯k
This contradicts (1.13).
Definition 1.21 Let Ω ⊂⊂ Cn be an open set. Ω is called an analytic polyhedron if there exist a neighborhood U of Ω and a finite number of functions f1 , · · · , fk ∈ O(U ) such that Ω = {z ∈ U | |f1 (z)| < 1, · · · , |fk (z)| < 1}. The collection of functions f1 , · · · , fk is called a frame for Ω. Theorem 1.17
Every analytic polyhedron is holomorphically convex.
Proof. Let Ω = {z ∈ U | |f1 (z)| < 1, · · · , |fk (z)| < 1}, where U is a neighborhood of Ω and f1 , · · · , fk ∈ O(U ). Let K ⊂ Ω be compact. We set rj = supK |fj |. Then rj < 1. Now we have b O ⊂ {z ∈ U | |f1 (z)| ≤ r1 , · · · , |fk (z)| ≤ rk } ⊂⊂ Ω, K Ω
which implies that Ω is holomorphically convex.
44
Several Complex Variables and Integral Formulas
Theorem 1.18 Let Ω ⊂ Cn be an open set and let K be a compact subset of Ω. Suppose K is O(Ω)-convex. Then K has a neighborhood basis consisting of analytic polyhedra defined by frames of functions holomorphic in Ω. O ∩ ∂U O, K Proof. Let U ⊂⊂ Ω be a neighborhood of K. Since K = K Ω Ω is empty. Let a ∈ ∂U . Then there exists fa ∈ O(Ω) such that |fa (a)| > supz∈K |fa (z)|. We choose r such that |fa (a)| > r > supz∈K |fa (z)|, and set ga = fa /r, then we have |ga (a)| > 1, supz∈K |ga (z)| < 1. Thus there exists a neighborhood Wa of a such that for z ∈ Wa , |ga (z)| > 1. By the argument of compactness, there exist open sets W1 , · · · , Wk and functions g1 , · · · , gk ∈ O(Ω) such that k
∂U ⊂ ∪ Wj , j=1
|gj (z)| > 1 (z ∈ Wj ).
We set Ω = {z ∈ U | |fj (z)| < 1, j = 1, · · · , k}. Then K ⊂ Ω ⊂⊂ U , which completes the proof of Theorem 1.18.
Exercises 1.1
Let f be a C 1 function defined on a domain in C. Show that the following equalities hold. ∂ f¯ ∂f = , ∂z ∂ z¯
1.2
∂ f¯ ∂f = . ∂ z¯ ∂z
Let Ω ⊂ Cn be an open set. Show that a real-valued function u on Ω is upper semicontinuous if and only if lim sup u(z) ≤ u(a) Ω∋z→a
where we define lim sup u(z) = lim Ω∋z→a
δ→0+
(a ∈ Ω),
sup z∈Ω∩B(a,δ)
u(z) .
1.3 (Maximum principle) Let Ω ⊂ Cn be a domain and let f be a holomorphic function in Ω. Suppose there exists a point ξ ∈ Ω such that |f (z)| ≤ |f (ξ)| for all z ∈ Ω. Show that f is constant. 1.4
Let Ω ⊂ Cn be an open set and let {fj } be a sequence of holomorphic functions in Ω. Suppose {fj } converges uniformly to f on every compact subset of Ω. Show that f is holomorphic in Ω.
Pseudoconvexity and Plurisubharmonicity
1.5
45
Let f be a holomorphic function in a domain Ω ⊂ Cn . Suppose there exists a point ξ ∈ Ω such that for all multi-indices α = (α1 , · · · , αn ), (∂ α f )(ξ) :=
∂ |α| f (ξ) = 0, ∂z1α1 · · · ∂znαn
where each αj is a nonnegative integer and |α| = α1 + · · · + αn . Show that f = 0. 1.6
Construct the function λ in Theorem 1.14.
1.7 (Schwarz lemma) Let f be a holomorphic function in the unit disc B(0, 1) ⊂ C. Assume that f (0) = 0 and |f (z)| ≤ 1 for z ∈ B(0, 1). Prove that |f (z) ≤ |z|,
|f ′ (0)| ≤ 1.
If either |f (z)| = |z| for some z 6= 0 or if |f ′ (0)| = 1, prove that f is expressed by f (z) = αz for some complex constant α of unit modulus. 1.8 (Schwarz-Pick lemma) Let f : B(0, 1) → B(0, 1) be a holomorphic function in the unit disc B(0, 1) ⊂ C. Assume that f (z1 ) = w1 and f (z2 ) = w2 for some z1 , z2 ∈ B(0, 1). Show that w1 − w2 z1 − z2 ≤ 1 − w1 w ¯2 1 − z1 z¯2 and
|f ′ (z1 )| ≤
1 − |w1 |2 . 1 − |z1 |2
If the equality holds one of the above inequalities, prove that f : B(0, 1) → B(0, 1) is a one-to-one onto mapping. 1.9
Let f : Ω → C and g : Ω → C be holomorphic functions in an open set Ω ⊂ C and let a ∈ Ω. If f (a) = g(a) = 0 and g ′ (a) 6= 0, prove that f ′ (z) f (z) = lim ′ . z→a g (z) z→a g(z) lim
1.10 (Uniqueness theorem) Let f : Ω → C be a holomorphic function in a domain Ω ⊂ C. If there exist a point a ∈ Ω and a sequence {zn } in Ω which converges a such that zn 6= a and f (zn ) = 0 for all n, then f = 0.
Several Complex Variables and Integral Formulas
46
1.11 (Open mapping theorem) Let f : Ω → C be a non-constant holomorphic function in an open set Ω ⊂ C. Prove that f (Ω) is an open set. 1.12
Let f be a holomorphic function in a simply connected domain Ω ⊂ C. Assume that f never vanishes. Prove the following:
(1) For a natural number m, there exists a holomorphic function g in Ω such that f = g m . (2) There exists a holomorphic function h in Ω such that f = eh . 1.13
Prove the following: Let f : Ω → C be a holomorphic function in an open set Ω ⊂ C. If f is one-to-one, then f ′ (z) = 0 for all z ∈ Ω.
1.14
Prove the following: Let f : Ω → C be a holomorphic function in a domain Ω ⊂ C. If f is one-to-one, then f −1 : f (Ω) → Ω is holomorphic. Moreover, (f −1 )′ (w) = {f ′ (f −1 (w))}−1 .
Chapter 2
The ∂¯ Problem in Pseudoconvex Domains In this chapter we give the proof of L2 estimates for the ∂¯ problem in ormander [HR2]. The assertion that pseudoconvex domains in Cn due to H¨ Ω pseudoconvex implies Ω is a domain of holomorphy is known as the Levi problem. The Levi problem was first solved affirmatively in C2 by Oka [OkA1] in 1942, and in Cn it was solved independently by Oka [OkA3], Bremermann [BRE] and Norguet [NOR] in the early 1950s. In 2.2 we give the proof of the Levi problem by the method of H¨ ormander [HR2]. In 2 2.3 we prove L extensions of holomorphic functions from submanifolds of bounded pseudoconvex domains in Cn which was first proved by Ohsawa and Takegoshi [OHT].
2.1
The Weighted L2 Space
For the preparation of the next section, we study the weighted L2 space whose element consists of differential forms. Moreover, we prove Green’s theorem which is useful for the proof of the Ohsawa-Takegoshi extension theorem. Definition 2.1 Let Ω ⊂⊂ Rn be a domain with C 1 boundary and let ρ be a defining funtion for Ω, that is, ρ is a real-valued C 1 function in a neighborhood G of Ω and satisfies
Ω = {x ∈ G | ρ(x) < 0},
n ∂ρ dρ(x) := (x)dxj = 0 ∂x j j=1
47
(x ∈ ∂Ω).
Several Complex Variables and Integral Formulas
48
Define the surface element dS by dS =
n j=1
(−1)j−1 νj dx1 ∧ · · · ∧ [dxj ] ∧ · · · ∧ dxn ,
(2.1)
where, [dxj ] means that dxj is omitted, and ν = (ν1 , · · · , νn ) is the unit outward normal vector for the boundary ∂Ω. If we set |dρ| =
"
∂ρ ∂x1
2
+ ···+
∂ρ ∂xn
2
,
then ν can be written 1 ν= |dρ|
∂ρ ∂ρ ,··· , ∂x1 ∂xn
.
Now we prove Green’s theorem. (Green’s theorem) Let u be a C 1 function on Ω. Then ∂u ∂ρ dS = u dV, Ω ∂xj ∂Ω ∂xj |dρ|
Theorem 2.1
where dV is the Lebesgue measure in Rn . Proof.
We set d[x]k = dx1 ∧ · · · ∧ [dxk ] ∧ · · · ∧ dxn .
Then by (2.1) we obtain
∂Ω
∂ρ dS = u ∂xj |dρ| =
∂Ω
∂Ω
+
n ∂ρ ∂ρ 1 (−1)k−1 u d[x]k 2 ∂xj |dρ| ∂xk k=1 ∂ρ ∂ρ 1 u d[x]k (−1)k−1 ∂xj |dρ|2 ∂xk
∂Ω
k=j
∂ρ ∂ρ 1 u (−1)j−1 d[x]j . ∂xj |dρ|2 ∂xj
Since ρ = 0 on ∂Ω, we have ∂ρ ∂ρ dxj = − dxi . ∂xj ∂xi i=j
The ∂¯ Problem in Pseudoconvex Domains
Consequently,
∂Ω
∂ρ dS u = ∂xj |dρ|
u(−1)j−1 d[x]j =
Ω
∂Ω
49
∂u dV, ∂xj
which completes the proof of Theorem 2.1.
Definition 2.2 Let Ω ⊂ Cn be an open set and let ϕ ∈ C ∞ (Ω) be a real-valued function. We denote by L2 (Ω, ϕ) the space of L2 integrable functions with respect to the measure e−ϕ dV , where dV is the Lebesgue measure in Cn . Let p and q be integers with 0 ≤ p, q ≤ n. For multiindices α = (i1 , · · · , ip ) and β = (j1 , · · · , jq ), where i1 , · · · , ip , j1 , · · · , jq are integers between 1 and n, define |α| = p, |β| = q and dz α = dzi1 ∧ · · · ∧ dzip ,
zjq . d¯ z β = d¯ zj1 ∧ · · · ∧ d¯
We also denote by L2(p,q) (Ω, ϕ) the space of all (p, q) forms f on Ω whose coefficients fα,β belong to L2 (Ω, ϕ). Definition 2.3
Let f be a (p, q) form in Ω. Then f is expressed by ′ f= fα,β dz α ∧ d¯ zβ , |α|=p
|β|=q
where ′ implies that the summation is performed only over strictly increasing multi-indices. Further, we set |f |2 =
α,β
′
|fα,β |2 .
By definition, f ∈ L2(p,q) (Ω, ϕ) means that f 2ϕ :=
Ω
|f |2 e−ϕ dV < ∞.
We denote by L2(p,q) (Ω, loc) the space of all (p, q) forms f on Ω whose coefficients fα,β are L2 functions on every compact subset of Ω. For f , g ∈ L2(p,q) (Ω, ϕ) with f=
α,β
′
fα,β dz α ∧ d¯ zβ ,
g=
α,β
′
gα,β dz α ∧ d¯ zβ ,
Several Complex Variables and Integral Formulas
50
we define the inner product of f and g by ′ (f, g) = fα,β gα,β e−ϕ dV. α,β
Ω
Then L2(p,q) (Ω, ϕ) is a Hilbert space. Definition 2.4
For g ∈ C 1 (Ω), define ∂ ∂g ∂ϕ (ge−ϕ ) = −g . ∂zj ∂zj ∂zj
δj g = e ϕ
In order to prove Theorem 2.2 we need the following lemma. Lemma 2.1 Let Ω be a bounded open set in Cn with C 1 boundary and let ρ be a defining function for Ω. For ′ ′ 1 1 (Ω), uI.K dz I d¯ z K ∈ C(p,q−1) (Ω), u = fI,J dz I d¯ z J ∈ C(p,q) f= I,K
I,J
we have ¯ f ) = (−1)p (∂u,
n ′ ∂uI,K Ω I,K j=1
= (−1)p−1
∂ z¯j
n ′
fI,jK e−ϕ dV
uI,K δj fI,jK e−ϕ dV
Ω I,K j=1
+(−1)p
′
∂Ω I,K
uI,K
n j=1
fI,jK
∂ρ −ϕ dS . e ∂zj |dρ|
Proof. We prove Lemma 2.1 in case p = 0, q = 1. The other cases will be left to the reader. We set zj = x2j−1 + ix2j . Then it follows from Green’s theorem that ∂ρ dS ∂u = u . ∂x ∂x j j |dρ| ∂Ω Ω If w is a C 1 function on Ω, then we obtain dS ∂ ∂ρ = uwe ¯ −ϕ (uwe ¯ −ϕ )dV ¯j |dρ| ¯j Ω ∂z ∂Ω ∂ z ∂u −ϕ uδj we−ϕ dV. we ¯ dV + = ¯j Ω Ω ∂z By setting w = fj and adding with respect to j, we obtain the desired equality.
The ∂¯ Problem in Pseudoconvex Domains
Definition 2.5
For f =
′
I,J fI,J dz
∂¯∗ f = (−1)p−1
I
n ′ I,K j=1
51
1 ∧ d¯ z J ∈ C(p,q) (Ω), define
δj fI,jK dz I ∧ d¯ zK .
f ∈ Def (∂¯∗ ) means that n
fI,jK
j=1
∂ρ =0 ∂zj
on ∂Ω.
for every multi-index I and K. For f ∈ Def (∂¯∗ ), it follows from Lemma 2.1 that ¯ f ) = (u, ∂¯∗ f ). (∂u, The following theorem was proved by H¨ ormander [HR1]. Theorem 2.2 Let Ω ⊂⊂ Cn be an open set with C 2 boundary and let ′ I z J be a C 2 (p, q) ρ be a defining function for Ω. Let α = I,J αI,J dz ∧ d¯ ∗ 2 ¯ form on Ω and let α ∈ Def (∂ ), ϕ a C function on Ω. Then ¯ 2= ∂¯∗ α + ∂α
n ′
I,K j,k=1 Ω n ′
αI,jK αI,kK
∂ 2 ϕ −ϕ e dV ∂zj ∂ z¯k
∂αI,J 2 −ϕ + ∂ z¯j e dV Ω I,J j=1 n ′ ∂ 2 ρ −ϕ dS . e αI,jK αI,kK + ∂zj ∂ z¯k |dρ| ∂Ω I,K j,k=1
Proof. We prove Theorem 2.2 in case p = 0, q = 1 and the other cases will be left to the reader. Let w be a C 2 function on Ω. Then we have ∂ ∂ ∂2ϕ − δk w = w . δk ∂ z¯j ∂ z¯j ∂zk ∂ z¯j
Several Complex Variables and Integral Formulas
52
Thus for C 2 functions v and w, we have ∂v ∂w −ϕ e dV δj vδk we−ϕ dV − ¯k ∂ z¯j Ω Ω ∂z ∂ 2 ϕ −ϕ dS ∂ρ vw ¯ e dV + vδk we−ϕ = ∂zk ∂ z¯j |dρ| Ω ∂Ω ∂zj ∂ρ ∂w −ϕ dS . v e − ∂ ¯k ∂ z¯j |dρ| ∂Ω z On the other hand we have ¯ = ∂α
n ∂αj ∂αk ∂αj zj , d¯ zk ∧ d¯ zj = − d¯ zk ∧ d¯ ∂ z¯k ∂ z¯k ∂ z¯j j>k
j,k=1
2 ∂αj ∂αk 2 ∂αj 2 ∂αk ∂αj ∂αk ∂αj ∂αk + − = ∂ z¯k ∂ z¯j ∂ z¯k − ∂ z¯j ∂ z¯k − ∂ z¯k ∂ z¯j . ∂ z¯j
Therefore we have
¯ 2= ∂¯∗ α2 + ∂α
n
j,k=1 n
δj αj δk αk e−ϕ dV
Ω
∂αj ∂αk −ϕ e dV ∂ z¯k ∂ z¯j j,k=1 Ω n ∂αk 2 −ϕ + ¯j e dV. Ω ∂z
−
j,k=1
Taking account of the boundary condition, we have
¯ 2 ∂¯∗ α2 + ∂α n n ∂αk 2 −ϕ ∂ 2 ϕ −ϕ e dV + α α e dV = j k ∂ z¯j ∂zj ∂ z¯k −
j,k=1 Ω n
j,k=1
∂Ω
j,k=1
αj
Ω
∂ρ ∂αk −ϕ dS . e ∂ z¯k ∂ z¯j |dρ|
By Lemma 1.21, there exists a C 1 function λ such that n
k=1
αk
∂ρ = λρ. ∂zk
The ∂¯ Problem in Pseudoconvex Domains
53
Hence we have on ∂Ω n ∂αk ∂ρ ∂2ρ ∂ρ + αk . =λ ∂ z¯j ∂zk ∂ z¯j ∂zk ∂ z¯j
k=1
If we multiply by αj and add with respect to j, we obtain on ∂Ω using the boundary condition n ∂αk ∂ρ ∂2ρ + αj αk αj = 0, ∂ z¯j ∂zk ∂ z¯j ∂zk j,k=1
which completes the proof of Theorem 2.2. 2.2
L2 Estimates in Pseudoconvex Domains
In this section we study L2 estimates for the ∂¯ problem in pseudoconvex domains in Cn by following H¨ormander [HR2]. In Chapter 1 we proved that every domain of holomorphy is a pseudoconvex domain. Here, we prove that every pseudoconvex domain is a domain of holomorphy by applying L2 estimates for solutions of the ∂¯ problem. Let H 1 and H 2 be Hilbert spaces. We denote the inner product of 1 H by (x, y)1 for x, y ∈ H 1 , and the inner product of H 2 by (x, y)2 for x, y ∈ H 2 . Let D ⊂ H 1 be a dense subset of H 1 and let T : D → H 2 be a linear operator. Then we set D = DT , T (D) = RT . Definition 2.6
Let T : D → H 2 be a linear operator. Define GT := {(x, T x) | x ∈ DT } ⊂ H 1 × H 2 .
We say that T is a closed operator if its graph GT is a closed subspace of H 1 × H 2. Definition 2.7 Let y ∈ H 2 . We say that y ∈ DT ∗ if there exists a constant c = c(y) > 0 such that |(T x, y)2 | ≤ cx1 for all x ∈ DT . By definition DT ∗ is a subspace of H 2 . Lemma 2.2
For y ∈ DT ∗ there exists a unique z ∈ H 1 such that (x, z)1 = (T x, y)2
Several Complex Variables and Integral Formulas
54
for all x ∈ DT . We set z = T ∗ y. Then T ∗ : DT ∗ → H 1 is a linear operator and satisfies (x, T ∗ y)1 = (T x, y)2
(2.2)
for all x ∈ DT , y ∈ DT ∗ . Proof.
For y ∈ DT ∗ , x ∈ DT , define ϕ(x) = (T x, y)2 .
Then ϕ is a linear functional on DT and satisfies |ϕ(x)| ≤ cx1 for some constant c. Hence ϕ is bounded. Since DT is dense in H 1 , for x ∈ H 1 there exists xν ∈ DT such that xν → x. We have |ϕ(xν ) − ϕ(xµ )| ≤ cxν − xµ → 0
(ν, µ → ∞).
Hence {ϕ(xν )} converges. If we define ϕ(x) := limν→∞ ϕ(xν ), then ϕ is a bounded linear functional on H 1 . By the Riesz representation theorem, there exists a unique z ∈ H 1 such that ϕ(x) = (x, z)1 for all x ∈ H 1 . Thus we have (x ∈ DT , y ∈ DT ∗ ).
(x, z)1 = (T x, y)2
Next we show that T ∗ is linear. For y1 , y2 ∈ DT ∗ and x ∈ DT , we have (x, T ∗ (y1 + y2 ))1 = (T x, y1 + y2 )2 = (x, T ∗ y1 + T ∗ y2 )1 . Since DT is dense in H 1 , we have T ∗ (y1 + y2 ) = T ∗ y1 + T ∗ y2 . Similarly, we have T ∗ (αy) = αT ∗ y for α ∈ C, y ∈ DT ∗ , which means that T ∗ is a linear operator. Lemma 2.3 Proof.
T ∗ : DT ∗ → H 1 is a closed operator.
It is sufficient to show that GT ∗ = {(y, T ∗ y) | y ∈ DT ∗ } ⊂ H 2 × H 1
is closed. Suppose (yn , zn ) ∈ GT ∗ ,
(yn , zn ) → (y0 , z0 ).
The ∂¯ Problem in Pseudoconvex Domains
55
Then yn ∈ DT ∗ and zn = T ∗ yn . Since {zn } is bounded, there exists a constant M > 0 such that zn < M for all n. For x ∈ DT , we have |(T x, yn )2 | = |(x, zn )1 | ≤ x zn ≤ M x1 . Letting n → ∞, we have |(T x, y0 )2 | ≤ M x1 , which means that y0 ∈ DT ∗ . On the other hand, we have |(x, T ∗ yn )1 − (x, T ∗ y0 )1 | = |(T x, yn − y0 )2 | ≤ T x2 yn − y0 2 → 0. Then (x, zn )1 → (x, T ∗ y0 )1 . Hence we have (x, z0 )1 = (x, T ∗ y0 )1
(x ∈ DT ).
Thus we have z0 = T ∗ y0 , and hence (y0 , z0 ) ∈ GT ∗ , which means that GT ∗ is closed. Lemma 2.4
Suppose DT ∗ is dense in H 2 . Then we have DT ∗∗ ⊃ DT ,
Proof.
T ∗∗ |DT = T.
For x ∈ DT and y ∈ DT ∗ , we have |(x, T ∗ y)1 | = |(T x, y)2 | ≤ T x2 y2 ,
which means that x ∈ DT ∗∗ . Thus DT ⊂ DT ∗∗ . On the other hand we have (T x, y)2 = (x, T ∗ y)1 = (T ∗ y, x)1 = (y, T ∗∗ x)2 = (T ∗∗ x, y)2 . Since DT ∗ is dense in H 2 , we obtain T x = T ∗∗ x for x ∈ DT , and hence T ∗∗ |DT = T . Lemma 2.5 Let T : DT → H 2 be a closed operator. Then DT ∗ is dense in H 2 and T ∗∗ = T . Proof. Define H = H 1 × H 2 . Let (x, y) ∈ H and (u, v) ∈ H. We define the inner product in H by < (x, y), (u, v) >:= (x, u)1 + (y, v)2 . Then H is a Hilbert space. Further, define J : H → H by J(x, y) = (−x, y). We define GT and GT∗ by GT := {(x, T x) | x ∈ DT } ⊂ H
56
Several Complex Variables and Integral Formulas
and GT∗ := {(T ∗ y, y) | y ∈ DT ∗ } ⊂ H. Then for y ∈ H 1 , z ∈ H 2 we have < (−x, T x), (y, z) >= 0 ⇐⇒ (x, y)1 = (T x, z)2
⇐⇒ z ∈ DT ∗ ,
∗
y = T z.
(x ∈ DT )
(x ∈ DT )
Hence we obtain (y, z)⊥JGT ⇐⇒ (y, z) ∈ GT∗ , which means that (JGT )⊥ = GT∗ . Since T is closed, GT is closed, and hence JGT is closed. Thus we have JGT = (GT∗ )⊥ . Similarly, we have GT = (JGT∗ )⊥ . Let u ∈ H 2 . Suppose (u, v) = 0 for all v ∈ DT ∗ . Since < (0, u), (T ∗ v, v) >= 0, we have (0, u) ∈ (GT∗ )⊥ . Hence (0, u) ∈ JGT . Thus there exists x ∈ DT such that (0, u) = (−x, T x), which implies that u = 0. Therefore if (u, v)2 = 0 for every v ∈ DT ∗ , then u = 0. If ϕ is a bounded linear functional on H 2 , then by the Riesz representation theorem, there exists z ∈ H 2 such that ϕ(v) = (v, z)2
(v ∈ H 2 ).
If ϕ = 0 on DT ∗ , then z = 0, which means that ϕ = 0 on H 2 . By applying the Hahn-Banach theorem, DT ∗ is dense in H 2 . Thus T ∗∗ : DT ∗∗ → H 2 is defined. We set GT∗∗ = {(z, T ∗∗z) | z ∈ DT ∗∗ } ⊂ H. Since T ∗ is closed, by using the same method as above we have (JGT∗ )⊥ = GT∗∗ . On the other hand, taking account of the equality (JGT∗ )⊥ = GT , we have GT = GT∗∗ , which means that DT = DT ∗∗ . It follows from Lemma 2.4 that T = T ∗∗ . Theorem 2.3 (Banach-Steinhaus theorem) Suppose X is a Banach space, Y is a normed linear space, and {Tα }α∈A is a collection of bounded linear operators of X into Y . Then either (1) or (2) holds:
The ∂¯ Problem in Pseudoconvex Domains
57
(1) There exists a constant M > 0 such that Tα ≤ M for every α ∈ A. (2) There exists a dense subset E of X such that sup Tα (x) = ∞
α∈A
for every x ∈ E. Proof.
We set ϕ(x) = sup Tα (x) α∈A
(x ∈ X)
and Vn = {x | ϕ(x) > n}. If we set fα (x) = Tα (x), then fα (x) is continuous, and hence Vn is an open set. Suppose VN is not dense in X. Then there exist x0 ∈ X and r > 0 such that if x ≤ r, then x0 + x ∈ VN . Thus ϕ(x0 + x) ≤ N . Hence we have Tα (x0 + x) ≤ N
(α ∈ A, x ≤ r),
which means that Tα (x) ≤ Tα (x0 + x) + Tα (x0 ) ≤ 2N. Then we obtain 1 2N Tα (rx) ≤ . r r x=1
Tα = sup Tα (x) = sup x=1
In case that all Vn are dense subset of X, by the Baire theorem, ∩∞ n=1 Vn is dense in X, and for x ∈ ∩∞ V we have ϕ(x) = ∞. n n=1 Theorem 2.4 Suppose D is a dense subspace in H 1 and T : D → H 2 is a closed operator. Let F be a closed subspace of H 2 and let F ⊃ RT . Then the following statements are equivalent: (a) F = RT .
Several Complex Variables and Integral Formulas
58
(b) There exists a constant c > 0 such that y2 ≤ cT ∗ y1 for all y ∈ F ∩ DT ∗ . Proof. (a) =⇒ (b). Suppose F = RT . Every element z ∈ H 2 is uniquely expressed by (z1 ∈ F, z2 ∈ F ⊥ ).
z = z1 + z2
Since z1 = T x for x ∈ DT , we have for y ∈ F ∩ DT ∗ |(y, z)2 | = |(y, z1 )2 | = |(y, T x)2 | = |(x, T ∗ y)1 | ≤ x1 T ∗ y1 .
(2.3)
We set K = {y | T ∗ y = 0, y ∈ DT ∗ } ∩ F. Further, we set for y ∈ K ϕy (z) =
(y, z)2 . T ∗ y1
Then by the Banach-Steinhaus theorem, either there exists a constant c > 0 such that ϕy < c
(2.4)
for every y ∈ K, or there exists a dense subset E of H 2 such that ∞ = sup |ϕy (z)|
(2.5)
y∈K
for every z ∈ E. By (2.3), (2.5) does not hold. Thus we have |(y, z)2 | 0 and c2 > 0 such that c1 ≤ e−ϕ(x) ≤ c2 Then
|f |2 dV ≤
K
which implies that f ∈ Lemma 2.9 If f ∈ f ∈ L2(p,q) (Ω, ϕ). Proof.
1 c1
K
(x ∈ K).
|f |2 e−ϕ(x) dV < ∞,
L2(p,q) (Ω, loc).
L2(p,q) (Ω, loc),
∞
then there exists ϕ ∈ C (Ω) such that
Suppose {Kn } is a sequence of compact subsets of Ω and that ◦
Kn ⊂⊂ K n+1 ⊂ Ω, We set
Kn
∞
∪ Kn = Ω.
n=1
|f |2 dV = cn .
We choose functions an ∈ Cc∞ (Cn ) with the properties 1 (z ∈ Kn − Kn−1 ) n 0 ≤ an (z) ≤ 1 (z ∈ C ), an (z) = . 0 (z = Kn+1 − Kn−2 ) For z ∈ Ω, define ϕ(z) =
∞
(log n2 (cn + 1))an (z).
n=1 ∞
Then ϕ ∈ C (Ω), and, ϕ(z) ≥ log n2 (cn + 1) (z ∈ Kn − Kn−1 ). Hence we have ∞ |f |2 e−ϕ dV |f |2 e−ϕ dV + |f |2 e−ϕ dV = Ω
K1
≤ c1 +
∞
n=2 ∞
n=2
Kn \Kn−1
Kn \Kn−1 2
|f |2 e− log n
= c1 +
1 c 2 (c + 1) n n n n=2
≤ c1 +
∞ 1 < ∞, n2 n=2
(cn +1)
dV
Several Complex Variables and Integral Formulas
62
which implies that f ∈ L2(p,q) (Ω, ϕ). Definition 2.9
For f ∈ L2(p,q) (Ω, ϕ), define ¯ = ∂f
n ∂fα,β ′ d¯ zk ∧ dzα ∧ d¯ zβ . ∂ z¯k
|α|=p
k=1
|β|=q
¯ exists in the sense of distributions. Then by Lemma 2.8, ∂f Definition 2.10 We denote by D(p,q) (Ω) the set of all C ∞ (p, q) forms in Ω whose supports are compact subsets of Ω. Further, we set D(Ω) = D(0,0) (Ω). Theorem 2.5 ¯ then ∂,
D(p,q) (Ω) is dense in L2(p,q) (Ω, ϕ1 ). Further, if we set T = T : DT → L2(p,q+1) (Ω, ϕ2 )
is a closed operator. Proof. Since D(p,q) (Ω) is dense in L2(p,q) (Ω, ϕ1 ), DT is dense in ¯ n ) → (f, g). L2(p,q) (Ω, ϕ1 ). We set GT = {(f, T f ) | f ∈ DT }. Suppose (fn , ∂f ¯ We set gn = ∂fn and ′ n ′ n gα,γ dz α ∧ d¯ zγ , fα,β dz α ∧ d¯ z β , gn = fn = α,γ
α,β
f=
α,β
′
fα,β dz α ∧ d¯ zβ ,
g=
α,γ
′
zγ . gα,γ dz α ∧ d¯
Then we have n gα,γ = (−1)p
′
ǫjβ γ
{j}∪β=γ
n ∂fα,β , ∂ z¯j
where ǫjβ γ means that if the permutation ρ which maps jβ to γ is even, jβ then ǫjβ γ equals 1, and ǫγ equals −1 if ρ is odd. For ψ ∈ D(Ω) we have n ∂ψ n ′ dV. (2.8) ǫjβ fα,β gα,γ ψdV = (−1)p−1 γ ∂ z¯j Ω Ω {j}∪β=γ
Since fn → f and gn → g, we have ∂ψ n ∂ψ fα,β dV → dV fα,β ∂ z ¯ ∂ z¯j j Ω Ω
The ∂¯ Problem in Pseudoconvex Domains
Ω
n gα,γ ψdV →
63
gα,γ ψdV.
Ω
Letting n → ∞ in (2.8) we have ∂ψ ′ dV, fα,β ǫjβ gα,γ ψdV = (−1)p−1 γ ∂ z¯j Ω Ω {j}∪β=γ
which means in the sense of distributions that ∂fα,β ′ ¯ )α,γ . ǫjβ gα,γ = (−1)p = (∂f γ ∂ z¯j {j}∪β=γ
¯ , and hence T is a closed operator. Thus we have g = ∂f
We set H 1 = L2(p,q) (Ω, ϕ1 ),
H 2 = L2(p,q+1) (Ω, ϕ2 ),
H 3 = L2(p,q+2) (Ω, ϕ3 ).
Further, we set ¯ ∈ H 2 }, D1 = {f ∈ H 1 | ∂f
¯ ∈ H 3 }, D2 = {f ∈ H 2 | ∂f
¯ D = T, ∂| 1
¯ D = S. ∂| 2
DT = D1 ,
DS = D2 .
Then
Lemma 2.10 Proof.
η ∈ D(Ω), f ∈ DS =⇒ ηf ∈ DS .
We have ¯ ) = η ∂f ¯ + ∂η ¯ ∧ f. ∂(ηf
Since the right side of (2.9) belongs to DS , we have ηf ∈ DS . Lemma 2.11 Proof.
f ∈ DT ∗ , η ∈ D(Ω) =⇒ ηf ∈ DT ∗ .
Let u ∈ DT . Using the equality (ηf, T u)2 = (f, η¯T u)2 = (f, T (¯ η u))2 − (f, ∂¯η¯ ∧ u)2 = (T ∗ f, η¯u)1 − (f, ∂¯η¯ ∧ u)2 ,
we have |(ηf, T u)2 | ≤ ηT ∗ f 1 u1 + f 2 ∂¯η¯ ∧ u2 .
(2.9)
Several Complex Variables and Integral Formulas
64
On the other hand, since supp(η) is compact, there exists a constant c > 0 such that |u|2 e−ϕ1 dV = cu21 . |∂¯η¯ ∧ u|2 e−ϕ1 eϕ1 −ϕ2 dV ≤ c ∂¯η¯ ∧ u22 = Ω
Ω
Hence we have |(ηf, T u)2 | ≤ (ηT ∗ f 1 +
√ cf 2 )u1 .
Thus ηf ∈ DT ∗ .
Lemma 2.12 Let Ω ⊂ Cn be an open set and let f be a nonnegative function in Ω. Suppose f is bounded on every compact subset of Ω. Then there exists a function ϕ ∈ C ∞ (Ω) such that f (z) ≤ ϕ(z) for z ∈ Ω. Proof. Let A = {Uν | ν = 1, 2, · · · } and B = {Vν | ν = 1, 2, · · · } be locally finite open covers of Ω such that Uν ⊂⊂ Vν ⊂⊂ Ω. Choose aν ∈ D(Ω) such that aν = 1 on U ν , supp(aν ) ⊂ Vν and 0 ≤ aν ≤ 1. We set supz∈Vν f (z) = Mν . Define ϕ(z) =
∞
Mν aν (z).
ν=1
Then ϕ satisfies the desired properties.
Lemma 2.13 Let Ω ⊂ Cn be an open set. Let {Kj }∞ j=0 be a sequence of compact subsets of Ω such that Kj−1 ⊂⊂ (Kj )◦ ,
∞
∪ Kj = Ω,
j=0
where (Kj )◦ denotes the interior of Kj . Let ηj ∈ D(Ω) be functions such that ηj = 1 on Kj−1 , supp(ηj ) ⊂ Kj ◦ and 0 ≤ ηj ≤ 1. Then there exists a function ψ ∈ C ∞ (Ω) such that n ∂ηj 2 ψ ∂ z¯k ≤ e
k=1
Proof.
Define f (z) =
2
∂ηj n k=1 ∂ z¯k 0
(j = 1, 2, · · · ).
(z ∈ Kj − Kj−1 , j = 1, 2, · · · ) (z ∈ K0 )
.
The ∂¯ Problem in Pseudoconvex Domains
65
Then f is bounded on every compact subset of Ω and satisfies n ∂ηj 2 f (z) ≥ (j = 1, 2, · · · ). ∂ z¯k k=1
By Lemma 2.12, there exists a function ψ ∈ C ∞ (Ω) such that f ≤ ψ in Ω. Since eψ(z) ≥ ψ(z), ψ satisfies the desired properties.
Definition 2.11 we set
Let ψ be the function in Lemma 2.13. For ψ ∈ C ∞ (Ω), ϕ1 = ϕ − 2ψ,
ϕ2 = ϕ − ψ,
ϕ3 = ϕ.
We assume ϕ ∈ C 2 (Ω). ϕ will be determined later. Then by Lemma 2.13 we have ¯ j |2 ≤ e−ϕ1 e−ϕ2 |∂η
¯ j |2 ≤ e−ϕ2 , e−ϕ3 |∂η
(j = 1, 2, · · · ).
Lemma 2.14 Let ηj , j = 1, 2, · · · , be functions in Lemma 2.13 and let f ∈ DS . Then S(ηj f ) − ηj S(f ) → 0 in H 3 as j → ∞. Proof. that
From the Schwarz inequality, there exists a constant c > 0 such
¯ j |2 |f |2 e−ϕ3 ≤ c|f |2 e−ϕ2 . ¯ j ∧ f |2 e−ϕ3 ≤ c|∂η |S(ηj f ) − ηj S(f )|2 e−ϕ3 = |∂η On the other hand, |S(ηj f ) − ηj S(f )|2 e−ϕ3 → 0 as j → ∞. Hence by the Lebesgue dominated convergence theorem, S(ηj f ) − ηj S(f )3 → 0. Lemma 2.15
Suppose f=
′
|α|=p
fα,β dz α ∧ d¯ z β ∈ DT ∗ .
|β|=q+1
Then we have ∗
p−1
T f = (−1)
n ′
|α|=p
|γ|=q
k=1
e
ϕ1
$ ∂ # −ϕ2 zγ , fα,kγ dz α ∧ d¯ e ∂zk
where we set for γ = (j1 , · · · , jq ) 0 (one of j1 , · · · , jq equals to k) . fα,kγ = (−1)r fα,δ (δ = (j1 , · · · , jr , k, jr+1 , · · · , jq ))
Several Complex Variables and Integral Formulas
66
Proof.
We set u=
′
|α|=p
uα,γ dz α ∧ d¯ z γ ∈ D(p,q) (Ω).
|γ|=q
Then we have n ∂uα,γ ′ d¯ zk ∧ dz α ∧ d¯ zγ ∂ z¯k
Tu =
|α|=p
k=1
|γ|=q
= (−1)p
′
|α|=p
|L|=q+1
′
εkK L
{k}∪K=L
∂uα,K α dz ∧ d¯ zL. ∂ z¯k
Since fα,J is defined to be skew-symmetric with respect to J, we have fα,L = εkK L fα,kK . Hence we have ′ (T ∗ f )α,γ uα,γ e−ϕ1 dV = (T ∗ f, u)1 = (f, T u)2 Ω
|α|=p
|γ|=q
p
= (−1)
Ω
= (−1)p
′
|α|=p
Σ′
{k}∪K=L
|L|=q+1
Ω
′
|α|=p
|L|=q+1
′
{k}∪K=L
εkK L
∂u ¯α,K fα,L e−ϕ2 dV ∂zk
∂u ¯α,K fα,kK e−ϕ2 dV ∂zk
n ∂u ¯α,K ′ fα,kK e−ϕ2 dV Ω ∂zk
= (−1)p
|α|=p
k=1
|K|=q p−1
= (−1)
n ′
|α|=p
k=1
eϕ1 Ω
|K|=q
$ ∂ # fα,kK e−ϕ2 u ¯α,K e−ϕ1 dV, ∂zk
which implies that ∗
p−1
(T f )α,γ = (−1)
n
k=1
e
ϕ1
$ ∂ # −ϕ2 fα,kγ . e ∂zk
Lemma 2.16 Let ηj , j = 1, 2, · · · , be functions in Lemma 2.13. Let f ∈ DT ∗ . Then T ∗ (ηj f ) − ηj T ∗ f 1 → 0
(j → 0).
The ∂¯ Problem in Pseudoconvex Domains
Proof.
67
Suppose f=
and
′
|α|=p
|β|=q+1
T ∗ (ηj f ) − ηj T ∗ f = It follows from Lemma 2.15 that
fα,β dz α ∧ d¯ zβ
′
|α|=p
|γ|=q
j gα,γ dz α ∧ d¯ zγ .
$ ∂ # −ϕ2 e ηj fα,kγ ∂zk k=1 n $ ∂ # −ϕ2 p−1 ϕ1 fα,kγ e e −(−1) ηj ∂zk
j gα,γ = (−1)p−1
n
eϕ1
k=1
= (−1)p−1
n
eϕ1
k=1
∂ηj −ϕ2 fα,kγ . e ∂zk
By Lemma 2.13 and the Schwarz inequality we have n n ∂ηj 2 j |gα,γ |2 ≤ e2(ϕ1 −ϕ2 ) |fα,kγ |2 ∂zk k=1 k=1 n ≤ eϕ1 −ϕ2 |fα,kγ |2 . k=1
Consequently, |T ∗ (ηj f ) − ηj T ∗ f |2 e−ϕ1 =
′
|α|=p
j |gα,γ |2 e−ϕ1
|γ|=q
≤
n ′
|α|=p
k=1
|γ|=q
= (q + 1)
|fα,kγ |2 e−ϕ2
′
|α|=p
|fα,β |2 e−ϕ2
|β|=q+1
= (q + 1)|f |2 e−ϕ2 . On the other hand, since |T ∗ (ηj f ) − ηj T ∗ f |2 e−ϕ1 converges to 0 almost everywhere, it follows from the Lebesgue dominated convergence theorem that T ∗ (ηj f ) − ηj T ∗ f 1 → 0 as j → 0.
Several Complex Variables and Integral Formulas
68
Definition 2.12
For f ∈ DT ∗ ∩ DS , we define f G := f 2 + T ∗ f 1 + Sf 3 .
Lemma 2.17 Let ηj , j = 1, 2, · · · , be functions in Lemma 2.13. Then for f ∈ DT ∗ ∩ DS ηj f − f G → 0
(j → ∞).
Proof. Since |ηj f − f | ≤ |f |, it follows from the Lebesgue dominated convergence theorem that ηj f − f 2 → 0. Similarly, we have ηj T ∗ f − T ∗ f 1 → 0. It follows from Lemma 2.16 that T ∗ (ηj f − f )1 = T ∗ (ηj f ) − ηj T ∗ (f )1 + ηj T ∗ f − T ∗ f 1 → 0 as j → ∞. Similarly, S(ηj f − f )3 → 0, and hence ηj f − f G → 0.
Lemma 2.18 Let f ∈ DS and supp(f ) ⊂⊂ Ω. Then for 0 < δ < 1, there exist fδ ∈ D(p,q+1) (Ω) such that fδ − f 2 → 0,
S(fδ ) − S(f )3 → 0
as δ → 0. Proof.
Choose a function Φ ∈ D(Cn ) such that ΦdV = 1, supp(Φ) ⊂⊂ B(0, 1). Cn
We set Φδ (z) = δ −2n Φ(z/δ). For a differential form f=
′
α,β
fα,β dz α ∧ d¯ sβ ,
define fδ =
α,β
′
(fα,β ∗ Φδ )dz α ∧ d¯ zβ ,
where fα,β ∗ Φδ (z) =
Cn
fα,β (z − ζ)Φδ (ζ)dζ.
The ∂¯ Problem in Pseudoconvex Domains
69
Then we have fδ ∈ D(p,q+1) (Ω) and fδ − f 2 → 0. On the other hand we have ∂fα,β ′ zk ∧ dz α ∧ d¯ ∗ Φδ d¯ zβ S(fδ ) = ∂ z¯k α,β ∂fα,K ′ ′ zL, = (−1)p εkK ∗ Φδ dz α ∧ d¯ L ∂ z¯k |α|=p
|L|=q+2
{k}∪K=L
which implies that S(fδ ) − S(f )3 → 0.
Lemma 2.19 Let f ∈ DT ∗ and supp(f ) ⊂⊂ Ω. Then there exist fδ ∈ D(p,q+1) (Ω) such that T ∗ (fδ ) − T ∗ (f )1 → 0 as δ → 0. Proof. n j=1
Since functions e
ϕ1
n ∂ −ϕ2 ∂fα,jγ ∂ϕ2 ϕ1 −ϕ2 fα,jγ ) = e (e − fα,jγ ∂zj ∂zj ∂zj j=1
are L2 functions and supp(f ) is a compact subset of Ω, gα,γ =
n ∂fα,jγ j=1
∂zj
−
∂ϕ2 fα,jγ ∂zj
are L2 functions. Thus, gα,γ ∗ Φδ − gα,γ L2 → 0. On the other hand we have gα,γ = (−1)p−1 eϕ2 −ϕ1 (T∗ f )α,γ . Therefore we obtain (−1)p−1 eϕ2 −ϕ1 T ∗ (fδ ) n ∂fα,jγ ∂ϕ2 ′ = ∗ Φδ − (fα,jγ ∗ Φδ ) dz α ∧ d¯ zγ ∂zj ∂zj α,γ j=1 ′ δ := ψα,γ dz α ∧ d¯ zγ . α,γ
Several Complex Variables and Integral Formulas
70
Then δ ψα,γ − gα,γ ∗ Φδ L2 (Ω) % % % % n n % % ∂ϕ2 ∂ϕ2 % ∗ Φ − f (f ∗ Φ ) =% δ α,jγ α,jγ δ % % ∂zj % 2 % j=1 ∂zj j=1 L (Ω) % % % % n % % ∂ϕ2 ∂ϕ2 fα,jγ ∗ Φδ − fα,jγ % ≤% % % ∂z ∂z j j % 2 % j=1 L (Ω) % % % % n % % ∂ϕ2 . (fα,jγ − fα,jγ ∗ Φδ ) % +% % % % %j=1 ∂zj L2 (Ω)
Consequently,
δ δ ψα,γ − gα,γ L2 (Ω) ≤ ψα,γ − gα,γ ∗ Φδ L2 (Ω) + gα,γ ∗ Φδ − gα,γ L2 (Ω) → 0,
which implies that T ∗ (fδ ) − T ∗ (f )1 → 0.
Theorem 2.6 For f ∈ DT ∗ ∩ DS , there exist fj ∈ D(p,q+1) (Ω), j = 1, 2, · · · , such that fj − f G → 0 as j → ∞. Proof.
For ε > 0, by Lemma 2.17 there exists j0 such that ηj0 f − f G
0 in Ω such that for z ∈ Ω and w ∈ Cn n
j,k=1
∂2Φ (z)wj w ¯k ≥ m(z)|w|2 . ∂zj ∂ z¯k
Define g(t) = max ρ(z) z∈Ωt
and h(t) = max z∈Ωt
2 ¯ 2(|∂ψ(z)| + eψ(z) ) m(z)
.
By Lemma 2.21 there exists a convex increasing function χ ∈ C ∞ (R) such that χ(t) ≥ g(t), χ′ (t) ≥ h(t). We set ϕ(z) = χ ◦ Φ(z). Then we have χ(Φ(z)) ≥ g(Φ(z)) = max ρ(w) ≥ ρ(z) w∈ΩΦ(z)
Several Complex Variables and Integral Formulas
74
and n
j,k=1
2 n ∂Φ ∂2ϕ ′′ (z)wj w ¯k = χ (Φ(z)) (z)wj ∂zj ∂ z¯k ∂z j=1 j +χ′ (Φ(z))
n
j,k=1
∂2Φ (z)wj w ¯k ∂zj ∂ z¯k
≥ h(Φ(z))m(z)|w|2 2 ¯ ≥ 2(|∂ψ(z)| + eψ(z) )|w|2 , which completes the proof of Theorem 2.7. Remark 2.1
We are going to prove the inequality f 22 ≤ T ∗ f 21 + Sf 23
(f ∈ DT ∗ ∩ .DS ).
(2.10)
If (2.10) holds, then we have for f ∈ F := Ker S f 2 ≤ T ∗ f 1
(f ∈ DT ∗ ∩ DS ).
Since RT ⊂ F ⊂ DS , we obtain f 2 ≤ T ∗ f 1
(f ∈ DT ∗ ∩ F ).
¯ = 0, then there By Theorem 2.4, we have F = RT , which implies that if ∂f 2 ¯ = f. exists u ∈ L(p,q) (Ω, ϕ1 ) such that ∂u Definition 2.13
For g ∈ C 1 (Ω), define δj g = e ϕ
∂g ∂ϕ ∂ (ge−ϕ ) = −g . ∂zj ∂zj ∂zj
Then for f ∈ C 2 (Ω) we obtain
∂f ∂ ∂2ϕ ∂ − (δj f ) = f . δj , f := δj ∂ z¯k ∂ z¯k ∂ z¯k ∂ z¯k ∂zj Theorem 2.8 Let Ω ⊂ Cn be a pseudoconvex domain and let ϕ ∈ C ∞ (Ω) be the function in Theorem 2.7. If we set ϕ1 = ϕ − 2ψ, ϕ2 = ϕ − ψ and ϕ3 = ϕ, then f 22 ≤ T ∗f 21 + Sf 23 for f ∈ D(p,q+1) (Ω).
The ∂¯ Problem in Pseudoconvex Domains
Proof.
For f ∈ D(p,q+1) (Ω), we have
T ∗ f = (−1)p−1
n ′
|α|=p
eϕ−2ψ
j=1
|γ|=q
= (−1)p−1 e−ψ
n ′
|α|=p
j=1
∂ −ϕ+ψ (e fα,jγ ) dz α ∧ d¯ zγ ∂zj
δj fα,jγ dz α ∧ d¯ zγ
|γ|=q
+(−1)p−1 e−ψ
n ′
|α|=p
fα,jγ
j=1
∂ψ α dz ∧ d¯ zγ ∂zj
|γ|=q
:= A + B. Then A21 = =
n ′
δj fα,jγ δk fα,kγ e−ϕ dV
Ω α,γ j,k=1
n ′
δj fα,jγ
Ω α,γ j,k=1 n ′
=−
Ω α,γ j,k=1
$ ∂ # fα,kγ e−ϕ dV ∂ z¯k
∂ (δj fα,jγ )fα,kγ e−ϕ dV. ∂ z¯k
Consequently, 2 n 2 n n ∂ψ ∂ψ ′ ′ ≤ fα,jγ |fα,jγ |2 ∂zj ∂z j α,γ j=1 α,γ j=1 j=1 n ′ = |∂ψ|2 |fα,jγ |2 . α,γ j=1
Hence we obtain
2T ∗ f 21 ≥ A21 − 2B21 n ∂ ′ (δj fα,jγ )fα,kγ e−ϕ dV =− ∂ ¯k Ω α,γ j,k=1 z n ′ |fα,jγ |2 e−ϕ dV. |∂ψ|2 −2 Ω
α,γ j=1
75
Several Complex Variables and Integral Formulas
76
On the other hand we have
Sf 23
n ∂fα,β 2 −ϕ ′ = ∂ z¯j e dV Ω α,β j=1
n ∂fα,kγ ∂fα,jγ ′ − e−ϕ dV. ∂ z¯k ∂zj Ω α,γ j,k=1
Using the equalities ∂fα,jγ −ϕ ∂ ∂ fα,kγ dV fα,jγ fα,kγ e−ϕ dV = e δj ∂ z¯k ∂ z¯k Ω ∂zj Ω ∂fα,jγ −ϕ ∂fα,kγ =− e dV, ¯k ∂zj Ω ∂z we have
Sf 23
n ∂fα,β 2 −ϕ ′ = ∂ z¯j e dV Ω α,β j=1
+
n ′
Ω α,γ j,k=1
∂ fα,jγ δj ∂ z¯k
fα,kγ e−ϕ dV.
Consequently,
2T
∗
f 21
+
Sf 23
n ∂ ∂ ′ ≥ − δj fα,jγ fα,kγ e−ϕ dV δj ∂ z¯k ∂ z¯k Ω α,γ j,k=1 n ′ −2 |fα,jγ |2 e−ϕ dV |∂ψ|2 Ω
α,γ j=1
n ∂2ϕ ′ fα,jγ fα,kγ e−ϕ dV = ¯k Ω α,γ j,k=1 ∂zj ∂ z n ′ |fα,jγ |2 e−ϕ dV. |∂ψ|2 −2 Ω
α,γ j=1
The ∂¯ Problem in Pseudoconvex Domains
77
It follows from Theorem 2.7 that n ′ ∗ 2 2 2T f 1 + Sf 3 ≥ |fα,jγ |2 e−ϕ dV 2(|∂ψ|2 + eψ ) Ω α,γ
−2 ≥ Corollary 2.1
Ω
j=1
n ′ |fα,jγ |2 e−ϕ dV |∂ψ|2 α,γ j=1
n ′
Ω α,γ j=1
2|fα,jγ |2 eψ−ϕ dV ≥ 2f 22.
For f ∈ DT ∗ ∩ DS , we have f 22 ≤ T ∗ f 21 + Sf 23.
Proof. For f ∈ DT ∗ ∩ DS , by Theorem 2.6 there exist fj ∈ D(p,q+1) (Ω) such that fj − f G → 0. On the other hand, by Theorem 2.8, fj 22 ≤ T ∗(fj )21 + S(fj )23 . Hence letting j → ∞, we have f 22 ≤ T ∗(f )21 + S(f )23 .
¯ = 0. Then there Theorem 2.9 Suppose f ∈ L2(p,q+1) (Ω, loc) satisfies ∂f ¯ = f. exists u ∈ L2(p,q) (Ω, loc) such that ∂u Proof. Let {Kj } be a sequence of compact subsets of Ω with the following properties: Kj ⊂⊂ (Kj+1 )◦ ,
∞
∪ Kj = Ω. (j = 1, 2, · · · ).
j=1
We set
Kj
|f |2 dV = Mj
(j = 1, 2, · · · ).
Let K0 = φ. Choose a function ϕ˜ ∈ C ∞ (Ω) such that ˜ e−ϕ(z)
0 such that X N ∂fε (x + θy)(xj − yj ) ≤ c3 kx − yk, |fε (x) − fε (y)| = j=1 ∂xj
which means that {fε } is equicontinuous. On the other hand, there exists a constant c4 > 0 such that
Z
∂N f ∂N
. fε (x) dV (x) ≤ c4 |fε (x)| ≤ ∂x1 · · · ∂xN L2 Rn ∂x1 · · · ∂xN
Hence {fε } are uniformly bounded. Using the Ascoli-Arzela theorem, one can choose a subsequence {fεj } of {fε } which converges uniformly to f˜ on every compact subset of Ω. Thus we have kfεj − f˜kL2 → 0. Since
Several Complex Variables and Integral Formulas
80
fεj − f L2 → 0, we have f = f˜ almost everywhere. Since f˜ is continuous, f is continuous almost everywhere. Theorem 2.10 Let f be a locally integrable function in RN with compact support. Suppose α ∂ f ∈ L2 (RN ) ∂x for all multi-indices α with |α| ≤ N + k + 1. Then there exists a function h ∈ C k (RN ) such that f = h almost everywhere. Proof. We prove Theorem 2.10 by induction on k. In case k = 0, Theorem 2.10 follows from Lemma 2.23. We assume that k ≥ 1 and Theorem 2.10 has already been proved for k − 1. For 1 ≤ j ≤ N , we have ∂f ∂α (|α| ≤ N + k). ∈ L2 (RN ) ∂xα ∂xj Hence by the inductive hypothesis, there exists f˜j ∈ C k−1 (RN ) such that ∂f ˜ ∂xj = fj almost everywhere. On the other hand, by Lemma 2.23 there exists a continuous function h such that f = h almost everwhere. Hence in ∂f ∂h = ∂x . Thus h is partially the sense of distributions, we obtain f˜j = ∂x j j ∂h ˜ differentiable and satisfies = fj for 1 ≤ j ≤ N , which means that ∂xj
h ∈ C k (RN ).
Corollary 2.2 Let Ω ⊂ RN be an open set. Suppose that f is a locally integrable function in Ω and that α ∂ (|α| = 0, 1, 2, · · · ). f ∈ L2loc (Ω) ∂x Then there exists a function f˜ ∈ C ∞ (Ω) such that f = f˜ almost everywhere. Proof.
Fix a function η ∈ D(Ω). We set ηf = h. Then we have α ∂ h ∈ L2 (RN ) (|α| = 0, 1, 2, · · · ). ∂x
˜ ∈ C(RN ) such that h = h ˜ It follows from Theorem 2.10 that there exists h almost everywhere. It follows from Theorem 2.10 that for every k there exists ϕ ∈ C k (RN ) such that h = ϕ almost everywhere. Thus, in the ˜ ∂ϕ ∂h ˜ is partially sense of distributions, we have ∂x = ∂x , which means that h j j ˜ ˜ ∈ C k (RN ). Since differentiable and satisfies ∂ h = ∂ϕ . Thus we have h ∂xj
∂xj
The ∂¯ Problem in Pseudoconvex Domains
81
k is arbitrary, ˜ h ∈ C ∞ (RN ). If we choose a sequence {Kn } of compact ◦
∞
subsets of Ω such that Kn ⊂ K n+1 , ∪ Kn = Ω, then f is of class C ∞ in n=1 Kn − An , where each set An is of Lebesgue measure 0. Since A = ∪∞ n=1 An , f is of class C ∞ almost everywhere.
Lemma 2.24 Let ϕ and ψ be as in Theorem 2.7. We set ϕ1 = ϕ − 2ψ, ¯ = 0, then there exists a ϕ2 = ϕ − ψ. If f ∈ L2(p,q+1) (Ω, ϕ2 ) satisfies ∂f unique u which satisfies T u = f,
u ∈ (Ker T )⊥ ,
u ∈ L2(p,q) (Ω, ϕ1 ).
Proof. First we show that Ker T is a closed subset. Let T uν = 0, uν → 0. Then (uν , T uν ) → (u, 0). By Theorem 2.5, T is a closed operator and satisfies u ∈ DT , (u, 0) = (u, T u). Hence T u = 0, which means that u ∈ Ker T . Consequently, Ker T is closed. Since Ker S = RT , there exists ¯ = f . Since Ker T is closed, α can be written α ∈ L2(p,q) (Ω, ϕ1 ) such that ∂α (α1 ∈ Ker T, α2 ∈ (Ker T )⊥ ).
α = α1 + α2 Define
Pα = α1 ,
α − Pα = u.
¯ = ∂α ¯ = f , which shows that u is the desired Then u ∈ (Ker T )⊥ and ∂u ∗ solution. Next we asuume u also satisfies the conditions of the lemma. Then u − u∗ ∈ (Ker T )⊥ , and T (u − u∗ ) = 0, which means that u − u∗ ∈ Ker T . Thus u = u∗ . Definition 2.15 Let Ω ⊂ Cn be an open set. For a nonnegative integer s, define the Sobolev space W s (Ω) of order s by µ η ∂ ∂ s 2 W (Ω) = f | f ∈ L (Ω), |µ| + |η| ≤ s . ∂z ∂ z¯ s Further, we define for f ∈ W(p,q) (Ω)
f 2W s (Ω)
=
′
|α|=p
|β|=q
Lemma 2.25
|µ|+|η|≤s
% µ η %2 % ∂ % ∂ % fα,β % % ∂z % 2 ∂ z¯
L (Ω)
For f ∈ D(Cn ), we have % % % % % % % ∂f % % = % ∂f % % (j = 1, 2, · · · , n). % ∂zj % 2 % ∂ z¯j % 2 L L
.
Several Complex Variables and Integral Formulas
82
Proof.
Using the integration by parts we have ¯ ∂f 2 ∂f ∂f ∂ f¯ ∂ dV = f dV = − dV ¯j ¯j ∂zj ¯j ∂zj Cn ∂ z Cn ∂ z Cn ∂ z ∂ f¯ ∂f ∂ f¯ ∂ dV f dV = =− ∂ z¯j ¯j Cn ∂zj ∂ z Cn ∂zj ∂f 2 dV. = Cn ∂zj
Lemma 2.26
Suppose that f ∈ L2 (Cn ) has a compact support and that ∂f ∈ L2 (Cn ) ∂ z¯j
(j = 1, · · · , n).
Then f ∈ W 1 (Cn ). Moreover, we have % % % % % % ∂f % % % % = % ∂f % % ∂zj % 2 % ∂ z¯j % 2 L L
Proof.
(j = 1, 2, · · · , n).
For ε > 0, we set fε = f ∗ Φε . Then fε ∈ D(Cn ). In L2 (Cn ) ∂fε ∂f ∂f = ∗ Φε → ∂ z¯j ∂ z¯j ∂ z¯j
(ε → 0).
It follows from Lemma 2.25 that % % % % % ∂fε % ∂fε ∂fδ % ∂fδ % % % % % % ∂zj − ∂zj % 2 = % ∂ z¯j − ∂ z¯j % 2 → 0. L L
Hence {∂fε /∂zj } is a Cauchy sequence. Thus there exists g ∈ L2 (Cn ) such that ∂fε →g ∂zj as ε → 0. For ψ ∈ D(Cn ) ∂fε ∂f ∂ψ ∂ψ , ψ = − fε , ,ψ , → − f, = ∂zj ∂zj ∂zj ∂zj which means that ∂f /∂zj = g. Hence we have ∂f /∂zj ∈ L2 (Cn ), and by Lemma 2.25 % % % % % % % ∂fε % % = % ∂fε % . % % ∂ z¯j % 2 % ∂zj % 2 L L
The ∂¯ Problem in Pseudoconvex Domains
Letting ε → 0, we obtain
Definition 2.16 define
% % % % % % % ∂f % % = % ∂f % . % % ∂ z¯j % 2 % ∂zj % 2 L L
For f ∈ L2(p,q+1) (Ω) with f =
T ∗ f := (−1)p−1
83
′
α,β
fα,β dz α ∧ d¯ zβ ,
n ∂fα,jγ α ′ dz ∧ d¯ zγ . ∂z j α,γ j=1
Lemma 2.27 Suppose that f is a differential form in L2(p,q+1) (Ω) with ∗ 2 ¯ ∈ L2 compact support and that ∂f (p,q+2) (Ω) and T f ∈ L(p,q) (Ω). Then 1 f ∈ W(p,q+1) (Ω). Proof. In case f ∈ D(p,q+1) (Ω), we set ψ = 0 and ϕ = 0 in the proof of Theorem 2.8 . Then n ∂fα,β 2 ′ ∗ 2 ¯ 2 (2.11) ∂ z¯j dV ≤ 2T f + ∂f . Ω α,β j=1
In the general case, setting f ∗ Φδ = fδ and applying (2.11) to fδ − fε , we obtain ¯ δ − ∂f ¯ ε 2 → 0 T ∗ fδ − T ∗ fε 2 + ∂f as ε, δ → 0. Consequently, n ∂(fα,β ∗ Φδ ) ∂(fα,β ∗ Φε ) 2 ′ dV → 0 − ∂ z¯j ∂ z¯j Ω j=1 α,β
as ε, δ → 0. Then there exists λα,β ∈ L2 (Ω) such that in L2 (Ω), ∂(fα,β ∗ Φδ ) → λα,β ∂ z¯j as δ → 0. Hence we have ∂fα,β /∂ z¯j = λα,β in the sense of distributions, and hence ∂fα,β /∂ z¯j ∈ L2 (Ω). By Lemma 2.26, ∂fα,β /∂zj ∈ L2 (Ω), and 1 (Ω). hence f ∈ W(p,q+1) Theorem 2.11 Let Ω ⊂ Cn be a pseudoconvex open set, and let 0 ≤ s ≤ s ¯ = 0, then there exists a solution u of (Ω, loc) and ∂f ∞. If f ∈ W(p,q+1) s ¯ ∂u = f such that u ∈ W(p,q) (Ω, loc).
Several Complex Variables and Integral Formulas
84
Proof. In case q = 0. By Theorem 2.9, there exists a solution u = ′ α 2 ¯ α uα dz of ∂u = f such that u ∈ L(p,q) (Ω, loc). Hence we have ∂uα = fα,j ∈ W s (Ω, loc). ∂ z¯j
θ (Ω, loc) for some θ with 0 ≤ θ ≤ s (if θ = 0, then Suppose that u ∈ W(p,0) θ u ∈ W(p,0) (Ω, loc)). For η ∈ D(Ω) we have
∂(ηuα ) ∂η = ηfα,j + uα ∈ W θ . ∂ z¯j ∂ z¯j For |µ| + |ν| ≤ θ, ∂ ∂ z¯j
∂ ∂zj
µ
∂ ∂ z¯j
ν
µ ν ∂ ∂ ∂ (ηuα ) ∈ L2 . (ηuα ) = ∂zj ∂ z¯j ∂ z¯j
By Lemma 2.26, ηuα ∈ W θ+1 . By the inductive hypothesis on θ, we have ηuα ∈ W s+1 . Thus we have u ∈ W s+1 (Ω, loc). In case q ≥ 1. Since f ∈ L2(p,q+1) (Ω, loc), by Lemma 2.9 there exists λ ∈ C ∞ (Ω) such that f ∈ L2(p,q+1) (Ω, λ). Suppose ϕ satisfies the condition of Theorem 2.7 for ρ(z) = ψ +λ. Then for ϕ1 = ϕ−2ψ and ϕ2 = ϕ−ψ, f ∈ L2(p,q+1) (Ω, ϕ2 ), and hence by Lemma 2.24, there exists u ∈ L2(p,q) (Ω, ϕ1 ) ¯ = f , u ∈ (Ker T )⊥ . Since RT = Ker S, RT is closed. By such that ∂u Lemma 2.6, RT ∗ is closed. By Lemma 2.7, u ∈ (Ker T )⊥ = RT ∗ = RT ∗ , which means that u = T ∗ v (v ∈ L2(p,q+1) (Ω, ϕ2 )). For v = Σ′ vα,β dz α ∧ d¯ zβ , α,β
we set
δv = (−1)p−1 T ∗ v =
n ∂fα,jγ ′
α,γ j=1
∂zj
dz α ∧ d¯ zγ .
Since e−ϕ1 u = (−1)p−1 δ(e−ϕ2 v), we obtain δ(e
−ϕ1
p−1
u) = (−1)
n n ∂(vα,jkL e−ϕ2 ) ′
α,L k=1 j=1
∂zk ∂zj
dz α ∧ d¯ z L = 0.
The ∂¯ Problem in Pseudoconvex Domains
85
Consequently, 0=
n ∂(e−ϕ1 uα,jγ ) ′
∂zj
α,γ j=1
dz α ∧ d¯ zγ
n n ∂uα,jγ α ∂ϕ1 ′ γ ′ = dz ∧ d¯ z − uα,jγ dz α ∧ d¯ zγ . ∂z ∂zj j α,γ j=1 α,γ j=1 θ Let 0 ≤ θ ≤ s. Suppose u ∈ W(p,q) (Ω, loc). When θ = 0, it is true. Fix η ∈ D(Ω). Then θ ¯ ∂(ηu) = η¯ ∧ u + ηf ∈ W(p,q+1) (Ω).
On the other hand we have (−1)p−1 T ∗ (ηu) n ∂(ηuα,jγ ) α ′ dz ∧ d¯ zγ = ∂z j α,γ j=1 =η
n ∂ϕ1 ′
α,γ j=1
∂zj
uα,jγ dz α ∧ d¯ zγ +
n ∂η ′ uα,jγ dz α ∧ d¯ zγ , ∂z j α,γ j=1
which implies that θ T ∗ (ηu) ∈ W(p,q−1) (Ω).
Suppose |µ| + |ν| ≤ θ. Then
µ ν ∂ ∂ (ηu) ∂z ∂ z¯ ν µ ∂ ∂ ′ α ¯ = (∂(ηu)) z γ ∈ L2(p,q+1) (Ω). α,γ dz ∧ d¯ ∂z ∂ z ¯ α,γ
∂¯
Similarly, T∗
∂ ∂z
µ
∂ ∂ z¯
ν
(ηu) ∈ L2(p,q−1) (Ω).
By Lemma 2.27 we have µ ν ∂ ∂ 1 (Ω), (ηu) ∈ W(p,q) ∂z ∂ z¯
Several Complex Variables and Integral Formulas
86
θ+1 which implies that ηu ∈ W(p,q) (Ω). By the inductive hypothesis on θ, s+1 s+1 (Ω, loc). ηu ∈ W(p,q) (Ω), and hence u ∈ W(p,q)
Corollary 2.3 Let Ω ⊂ Cn be a pseudoconvex domain. Suppose f is a ¯ = 0. Then there exists a C ∞ (p, q) form u C ∞ (p, q + 1) form in Ω with ∂f ¯ in Ω such that ∂u = f .
∞ ∞ Proof. Since C ∞ (Ω) ⊂ Wloc (Ω, loc). Then (Ω), we have f ∈ W(p,q+1) ∞ ¯ = f . By by Theorem 2.11, there exists u ∈ W(p,q) (Ω, loc) such that ∂u ˜ on Ω such that u = u˜ almost Corollary 2.2, there exists a C ∞ (p, q) form u everywhere. Thus we have ∂¯u ˜ = f.
Definition 2.17 A metric space is called separable if it has a countable everywhere dense subset. Definition 2.18 Let H be a Hilbert space. We say that a sequence {xn } in H converges weakly to x ∈ H if (xn , y) → (x, y) for every y ∈ H. Lemma 2.28 Every bounded sequence in a separable Hilbert space contains a weakly convergent subsequence. Proof. Suppose {xn } is a bounded sequence in a separable Hilbert space H. Then there exists a constant M such that xn ≤ M for all n. Since H is separable, there is a countable dense subset {zn }. We have |(xn , z1 )| ≤ xn z1 ≤ M z1 . Thus {(xn , z1 )} is bounded. Using Bolzano-Weierstrass theorem, there is a (1) (1) subsequence {xn } of {xn } such that {(xn , z1 )} converges. Since (1) |(x(1) n , z2 )| ≤ xn z2 ≤ M z2 , (1)
(2)
(2)
there exists a subsequence {xn } of {xn } such that {(xn , x2 )} converges. (k) Repeating this process, we have sequences {xn } such that (k+1)
(k)
(1) {xn } is a subsequence of {xn }. (k) (2) {(xn , zj )} converge for j = 1, · · · , k. (n)
(n)
Thus {(xn , z)} converge for z = z1 , z2 , · · · . We set yn = xn , and for ε > 0, we set δ = min{ε/(3M ), ε/3}. If w1 , w2 ∈ H, w1 − w2 < δ, then ε |(yn , w1 ) − (yn .w2 )| ≤ yn w1 − w2 < . 3
The ∂¯ Problem in Pseudoconvex Domains
87
Since for any z ∈ H, there exists zn0 such that z − zn0 < δ. Hence we have |(ym , z) − (yn , z)| ≤ |(ym , z) − (ym , zn0 )| + |(ym , zn0 ) − (yn , zn0 )| +|(yn , zn0 ) − (yn , z)| ε ε ≤ + |(ym , zn0 ) − (yn , zn0 )| + . 3 3
Since {(yn , zn0 )} converges, there exists a positive integer N such that if m, n ≥ N , then |(ym , zn0 ) − (yn , zn0 )|
α, then ϕ = ϕ1 on Ωα . Hence we have |uaj |2 e−ϕ dV ≤ 2A2 . Ωα
Using the equality (uaj , u)Hα = (uaj − u, u)Hα + (u, u)Hα , we obtain √ 2AuHα + |(uaj − u, u)Hα |, √ which implies that uHα ≤ 2A. Namely, for any α, |u|2 e−ϕ dV ≤ 2A2 . u2Hα ≤
Ωα
For f ∈ DT ∗ , we have (g, f )2 = (T uaj , f )2 = (uaj , T ∗ f ) → (u, T ∗ f ) = (T u, f ), which implies that T u = g.
Now we are going to prove L2 estimates for the ∂¯ problem in pseudoconvex domains obtained by H¨ ormander [HR2]. Theorem 2.13 (L2 estimates) Let Ω ⊂ Cn be a pseudoconvex open set and let ϕ be a plurisubharmonic function in Ω. If g ∈ L2(p,q+1) (Ω, ϕ) satisfies ¯ = 0, then there exists a solution u ∈ L2 (Ω, loc) of the equation ∂u ¯ =g ∂g (p,q) such that 2 −ϕ 2 −2 |g|2 e−ϕ dV. (2.14) |u| e (1 + |z| ) dV ≤ Ω
Ω
Several Complex Variables and Integral Formulas
92
Proof. Then
First we assume that ϕ ∈ C 2 (Ω). We set ϕ′ = ϕ + 2 log(1 + |z|2 ).
n ∂2 ∂ 2 ϕ′ wj w ¯k ≥ 2 {log(1 + |z|2 )}wj w ¯k ∂zj ∂ z¯k ∂zj ∂ z¯k j,k=1 j,k=1 n 2 |w|2 (1 + |z|2 ) − | z¯j wj |2 = (1 + |z|2 )2 j=1 n
≥ 2(1 + |z|2 )−2 |w|2 .
Then (2.14) follows from Theorem 2.12. In the general case, there exists a C ∞ strictly plurisubharmonic function ρ in Ω such that for any real number a Ωa = {z ∈ Ω | ρ(z) < a} ⊂⊂ Ω. depending only on |z1 |, · · · , |zn | and Suppose that Φ ∈ D(Cn ) is a function ! that Φ = 0 on |z| ≥ 1, 0 ≤ Φ and ΦdV = 1. Define ϕ(z − εζ)Φ(ζ)dV (ζ). ϕε (z) = Ω
Then ϕε is a C ∞ plurisubharmonic function in {z ∈ Cn | d(z, Ωc ) > ε}, and ϕε ↓ ϕ if ε ↓ 0. We choose a(ε) with the properties that a(ε) → ∞ if ε → 0, and that ϕε is a C ∞ plurisubharmonic function in Ωa(ε) . By using the result of the C 2 case, there exists uε ∈ L2(p,q) (Ωa(ε) , ϕε ) such that ¯ ε = g in Ωa(ε) , and ∂u |uε |2 e−ϕε (1 + |z|2 )−2 dV ≤ |g|2 e−ϕ dV. |g|2 e−ϕε dV ≤ Ωa(ε)
Ωa(ε)
Ω
Fix a real number α. We choose δ > 0 such that a(δ) > α. Then there exists a constant c1 > 0 such that c1 ≤ e−ϕδ (1 + |z|2 )−2
(z ∈ Ωα ),
which implies that for ε > δ, 1 2 |uε | dV ≤ |g|2 e−ϕ dV. c1 Ω Ωα Hence we can choose a subsequence {uεj } of {uε } which converges weakly in L2(p,q) (Ωα ). Let {αk } be a sequence such that αk → ∞. Since we can
The ∂¯ Problem in Pseudoconvex Domains
93
choose a subsequence {uj,k } of {uεj } which converges weakly in {Ωαk }, {uk,k } converges weakly to u in L2(p,q) (Ω, loc). Hence we have for ε > δ, 2 −2 2 −ϕε |g|2 e−ϕ dV. (1 + |z| ) dV ≤ |u| e Ω
Ωα
Letting ε → 0 we have |g|2 e−ϕ dV. |u|2 e−ϕ (1 + |z|2 )−2 dV ≤ Ω
Ωα
¯ = g, and α is arbitrary, we have (2.14). Since ∂u Theorem 2.14
Let Ω ⊂ Cn be a pseudoconvex open set. Define ω = Ω ∩ {zn = 0}.
Suppose f : ω → C is holomorphic in ω ˜ , where ω ˜ = {(z1 , · · · , zn−1 ) ∈ Cn−1 | (z1 , · · · , zn−1 , 0) ∈ ω}. Then there exists a holomorphic function F : Ω → C such that F |ω = f . Proof.
Let π : Cn → Cn−1 be the projection such that π(z1 · · · , zn ) = (z1 · · · , zn−1 ).
If B = {z ∈ Ω | π(z) ∈ ω ˜ }, then, ω and B are closed subset of Ω, and ω ∩ B = φ. Then there exists a function ψ ∈ C ∞ (Ω) such that ψ = 1 on an open subset of Ω which contains ω, and ψ = 0 on an open subset of Ω which contains B. Define F (z) = ψ(z)f (π(z), 0) + zn v(z), ¯ = 0. where v is a C ∞ function in Ω. We will determine v later to satisfy ∂F ∞ Then F is a C function on Ω and ¯ ¯ (z) = ∂ψ(z)f ¯ ∂F (π(z), 0) + zn ∂v(z). If v satisfies ¯ (−∂ψ(z))f (π(z), 0) ¯ , ∂v(z) = zn
(2.15)
¯ . Since ∂ψ(z) ¯ then 0 = ∂F = 0 in a neighborhood of ω, the right side of ∞ (2.15) is of class C in Ω. Further, the right side of (2.15) is ∂¯ closed. Therefore, by Corollary 2.3 there exists v ∈ C ∞ (Ω) which satisfies (2.15). Thus F is holomorphic in Ω. Since F |ω = f , F is the desired function.
94
Several Complex Variables and Integral Formulas
Theorem 2.15 Let Ω ⊂ Cn be an open set. Suppose that for every ∞ ¯ = 0, there exists u ∈ C ∞ (Ω) such ( 0 ≤ q ≤ n − 2) with ∂f f ∈ C(0,q+1) (0,q) ¯ = f . Then Ω is a domain of holomorphy. that ∂u Proof. We prove Theorem 2.15 by induction on n. If n = 1, then the theorem is true since every open set is a domain of holomorphy. Assume that the theorem has already been proved for n − 1 dimensions. In order to prove that the domain Ω ⊂ Cn is a domain of holomorphy, it is sufficient to show that for every open convex set D ⊂ Ω such that some boundary point z 0 of D is on ∂Ω, there is a holomorphic function in Ω which cannot be continued holomorphically to a neighboorhood of z 0 . (For, if a holomorphic function in Ω can be extended holomorphically to a neighborhood G of w ∈ ∂Ω, then there exists r > 0 such that B(w, 2r) ⊂ G. For z1 ∈ B(w, r) ∩ Ω, there exists δ(0 < δ ≤ r) such that B(z1 , δ) ⊂ Ω, ∂B(z1 , δ) ∩ ∂Ω = φ. Then every holomorphic function on Ω can be extended to a neighborhood of ∂B(z1 , δ) ∩ ∂Ω, which is a contradiction.) We choose a coordinate system such that z0 = 0 and D0 = {zn = 0} ∩ D is not empty. Since D is convex, 0 is a boundary point of D0 , and hence a boundary point of ω = ∞ ¯ = 0. For ω, let (ω) and ∂f {z | z ∈ Ω, zn = 0}. Suppose f ∈ C(0,q+1) ψ is the function in the proof of Theorem 2.14. Let i : ω → Ω be the inclusion mapping. Let π(z1 , · · · , zn ) = (z1 , · · · , zn−1 ). For z ∈ Ω and ∞ (Ω), we set v ∈ C(0,q+1) F (z) = ψ(z)π ∗ f (z) − zn v(z). ∞ Then F ∈ C(0,q+1) (Ω). We have
¯ ¯ = ∂ψ ¯ ∧ π ∗ f − zn ∂v. ∂F Hence, if we choose v such that ∗ ¯ ¯ = ∂ψ ∧ π f , ∂v zn
(2.16)
¯ = 0. Since the right side of (2.16) belongs to C ∞ then ∂F (0,q+1) (Ω), and ∞ ¯ ∂ closed, by the assumption, there exists v ∈ C(0,q) (Ω) satisfying (2.16). ∞ ¯ = 0 and i∗ F = f . By (Ω) such that ∂F Thus there exists F ∈ C(0,q+1) ∞ ¯ = F . If we set the assumption, there exists U ∈ C(0,q) (Ω) such that ∂U u = i∗ U , then u ∈ C(0,q) (ω), and ¯ = ∂i ¯ ∗ U = i∗ ∂U ¯ = i∗ u = f. ∂u
The ∂¯ Problem in Pseudoconvex Domains
95
By the inductive hypothesis, ω is a domain of holomorphy. Hence there exists a holomorphic function h in ω which cannot be extended holomorphically to a neighborhood of π(z0 ). On the other hand, by Theorem 2.14 there exists a holomorphic function H on Ω such that H|ω = h. Since H cannot be extended holomorphically in a neighborhood of z0 , Ω is a domain of holomorphy. Hence Theorem 2.15 holds for n. Corollary 2.4 (Levi’s problem) Let Ω ⊂ Cn be an open set. Then Ω is pseudoconvex if and only if Ω is a domain of holomorphy. Proof. Corollary 2.4 follows from Corollary 1.6, Corollary 2.3 and Theorem 2.15. Lemma 2.29 Suppose that p is a C ∞ strictly plurisubharmonic function in Ω and that for every c ∈ R Kc = {z ∈ Ω | p(z) ≤ c} ⊂⊂ Ω. Then every holomorphic function in a neighborhood of K0 can be approximated in L2 norm on K0 by functions in O(Ω). Proof. Let u be a holomorphic function in a neighborhood of K0 . By the Hahn-Banach theorem, it is sufficient to show that if ϕ is a bounded linear functional on L2 (K0 ) which vanishes on O(Ω), then ϕ(u) = 0. By the Riesz representation theorem, there exists v ∈ L2 (K0 ) such that x¯ v dV. ϕ(x) = (x, v) = K0
Hence it is sufficient to show that f v¯dV = 0 (f ∈ O(Ω)) =⇒
u¯ vdV = 0.
K0
K0
By setting v = 0 outside of K0 , we extend v to the function on Ω. If f ¯ = 0 on Ω, then f is holomorphic in Ω. Hence, by satisfies the equation ∂f the assumption, f v¯dV = 0, f v¯eϕ1 e−ϕ1 dV = (f, veϕ1 )1 = K0
K0
which means that veϕ1 ∈ (Ker T )⊥ . Suppose RT = Ker S. Then Ker S is closed, and hence RT is closed. By Lemma 2.6, RT ∗ is closed. By Lemma
Several Complex Variables and Integral Formulas
96
2.7, we have (Ker T )⊥ = RT ∗ , and hence veϕ1 ∈ RT ∗ = T ∗ (DT ∗ ). By Theorem 2.4, we have f 2 ≤ cT ∗ f 1
(f ∈ RT ∩ DT ∗ ).
Next we show that T ∗ (RT ∩ DT ∗ ) = T ∗ (DT ∗ ). Let u ∈ DT ∗ . Then u can be uniquely expressed by (u1 ∈ RT ∩ DT ∗ , u2 ∈ (RT )⊥ ∩ DT ∗ ).
u = u1 + u2
Since (RT )⊥ = Ker T ∗ , we obtain T ∗ (u) = T ∗ (u1 + u2 ) = T ∗ u1 . Hence we have T ∗ (RT ∩ DT ∗ ) = T ∗ (DT ∗ ), which means that there exists f ∈ RT ∩ DT ∗ such that veϕ1 = T ∗ f, that is, if f =
n
j=1
f 2 ≤ cT ∗f 1 ,
fj d¯ zj , then we have veϕ1 = −eϕ1
n ∂(e−ϕ2 fj ) j=1
∂zj
.
We set g = f e−ϕ2 . Then we have (1) v = − (2)
n Ω j=1
n ∂gj j=1
∂zj
|gj |2 eϕ2 dV ≤ c2
Define ρν (t) =
Ω
|v|2 eϕ1 dV .
2
νe−1/t (t > 0) , 0 (t ≤ 0)
λν (t) =
t
ρν (t)dt.
−∞
Let χ be the function in the proof of Theorem 2.7. Define χν = χ + λν , and ϕ1 = χν ◦ p − 2ψ,
ϕ2 = χν ◦ p − ψ,
ϕ3 = χν ◦ p,
where ψ is the function defined in Lemma 2.13. By Corollary 2.1, Ker S = RT . Hence for ϕi (i = 1, 2, 3) we can construct gν satisfying (1), (2). χν
The ∂¯ Problem in Pseudoconvex Domains
97
satisfies that χν (t) = χ(t) for t ≤ 0, and χν (t) ↑ ∞ (ν → ∞) for t > 0. Using (2), we have n Ω j=1
|gjν |2 exp(χν ◦ p − ψ)dV ≤ c2 = c2
Ω
|v|2 exp(χν ◦ p − 2ψ)dV
K0
|v|2 exp(χ ◦ p − 2ψ)dV
≤ C1 . Thus we obtain n Ω j=1
|gjν |2 exp(χ1 ◦ p − ψ)dV ≤ C1 ,
which means that {g ν } is a bounded sequence in L(0,1) (Ω, ψ −χ1 ◦p). Hence we can choose a subsequence {g νk } of {g ν } which converges weakly. Let g νk → g. For s2 > s1 > 0, we set M = max ψ(x). x∈Ks2
Since exp(χν ◦ p − ψ) ≥ exp(χν (s1 ) − M ) on Ks2 − Ks1 , we have
n
Ks2 −Ks1 j=1
|gjν |2 dV ≤ C1 exp(M − χν (s1 )) → 0 (ν → ∞).
Thus {gjν } converges 0 in L2 (Ks2 − Ks1 ), which means that {gjν } converges to 0 in Ω − K0 in the sense of distributions. Hence, g = 0 on Ω − K0 . v is written in the following form v=−
n ∂gjν j=1
∂zj
.
Letting ν → ∞ we have in the sense of distributions v=− Hence we have for u ∈ D(Ω)
n ∂gj j=1
∂zj
.
n n ∂¯ gj ∂u dV = g¯j dV. u u¯ vdV = − ∂ z ¯ ∂ ¯j j Ω j=1 z Ω Ω j=1
98
Several Complex Variables and Integral Formulas
Since g = 0 outside of K0 , we have for every holomorphic function in a neighborhood of K0 n ∂u u¯ v dV = g¯j dV = 0, ∂ ¯j Ω K0 j=1 z which completes the proof of Lemma 2.29.
Theorem 2.16 Let Ω ⊂ Cn be a pseudoconvex open set and let K be a ˆ P . Then there exists a function compact subset of Ω, ω a neighborhood of K Ω ∞ u ∈ C (Ω) such that (a) u is strictly plurisubharmonic in Ω. (b) u < 0 in K, u > 0 in Ω ∩ ω c . (c) {z ∈ Ω | u(z) < c} ⊂⊂ Ω for every c ∈ R. Proof. By Theorem 1.15, there exists a C ∞ strictly plurisubharmonic function u0 on Ω such that for every real number c, {z ∈ Ω | u0 (z) < c} ⊂⊂ Ω. Without loss of generality, we may assume that u0 < 0 in K. Define K ′ = {z ∈ Ω | u0 (z) ≤ 2},
L = {z | z ∈ Ω ∩ ω c , u0 (z) ≤ 0}.
ˆ P for each z ∈ L, Then L ⊂ K ′ , and K ′ and L are compact. Since z ∈ K Ω there exists g ∈ P (Ω) such that |g(z)| > sup g. K
Let d be such that |g(z)| > d > supK g. We set g˜ = g − d. Then g˜(z) > 0, g˜ < 0 in K. There exists a C ∞ plurisubharmonic function gε (z) defined in an open set W with K ′ ⊂ W ⊂ Ω such that gε ↓ g˜ (ε → 0). For any sufficiently small ε > 0, gε > 0 in some neighborhood of z, gε < 0 in K. Since L is compact, by the Heine-Borel theorem, there exist finitely many C ∞ strictly plurisubharmonic functions ϕ1 , · · · , ϕk in W such that if we define ϕ = max(ϕ1 , · · · , ϕk ), then ϕ is a continuous plurisubharmonic function in W , ϕ > 0 in some neighborhood of L and ϕ < 0 in K. We set c = sup ϕ > 0. K′
Define v(z) =
sup(ϕ(z), cu0 (z)) (u0 (z) < 2) . (u0 (z) > 1) cu0 (z)
The ∂¯ Problem in Pseudoconvex Domains
99
If 1 < u0 (z) < 2, then z ∈ K ′ , and hence ϕ(z) ≤ c < cu0 (z). Then sup(ϕ(z), cu0 (z)) = cu0 (z), which implies that v is a continuous plurisubharmonic function in Ω. Further, v < 0 in K. If u0 (z) ≤ 0 for z ∈ Ω ∩ ω c , then z ∈ L, and hence v(z) = ϕ(z) > 0. If u0 (z) > 0, then v(z) ≥ cu0 (z) > 0, which implies that v(z) > 0 for z ∈ Ω ∩ ω c . Thus v is a continuous plurisubharmonic function in Ω satisfying (b) and (c). We set Ωc = {z ∈ Ω | v(z) < c}, and vj (z) =
Ωj+1
v(ζ)λ((z − ζ)/ε) dV (ζ) + ε|z|2 , ε2n
where λ is the function defined in Theorem 1.14. If we choose ε > 0 sufficiently small, then vj ∈ C ∞ (Cn ). Further, vj > v, and vj is strictly ¯ j . Choosing ε sufficiently small, plurisubharmonic in a neighborhood of Ω we may assume that in K, v0 < 0 and v1 < 0. Further, vj < v + 1 in Ωj (j = 1, 2, · · · ). We choose a convex function χ ∈ C ∞ (R) such that χ(t) = 0 for t < 0, χ′ (t) > 0 for t > 0. Then χ(vj +1−j) is strictly plurisubharmonic in a neighborhood of Ωj \Ωj−1 . We choose aj (j = 1, 2, · · · ) such that if we set um = v0 +
m j=1
aj χ(vj + 1 − j),
then um is strictly plurisubharmonic in a neighborhood of Ω, and um > v. If i > j, k > j, then uk = ui in Ωj , which implies that u = limi→∞ ui exists. Since u is a C ∞ plurisubharmonic function in Ω, u = v0 < 0 in K and u > v in Ω, u satisfies (a), (b) and (c). Theorem 2.17 Let Ω ⊂ Cn be a pseudoconvex open set and K a comˆ P . Then every holomorphic function in a pact subset of Ω with K = K Ω neighborhood of K can be approximated uniformly on K by functions in ˆ O , Theorem 2.17 holds for every compact set K with ˆP ⊂ K O(Ω). (Since K Ω Ω O ˆ K = KΩ .) Proof. Let u be holomorphic in a neighborhood ω of K. By Theorem 2.16, there exists a C ∞ strictly plurisubharmonic function p in Ω such that p satisfies the assumption in Lemma 2.29, and if we set Kc = {z ∈ Ω | p(z) < c}, p satisfies K ⊂ K0◦ ⊂ K0 ⊂ ω. By Lemma 2.29, there exist uj ∈ O(Ω)
Several Complex Variables and Integral Formulas
100
such that uj → u in L2 (K0 ). Using Corollary 1.3, uj − u converges to 0 uniformly on K. 2.3
The Ohsawa-Takegoshi Extension Theorem
Let Ω ⊂⊂ Cn be a pseudoconvex domain and let H = {z ∈ Cn | zn = 0}. Then Ohsawa and Takegoshi [OHT] proved that every L2 holomorphic function in H ∩ Ω can be extended to an L2 holomorphic function in Ω. The proof given here is based on the proof of Jarnicki-Pflug [JP]. Let H j , j = 0, 1, 2, be Hilbert spaces. Let Dj be dense subsets of H j , j = 0, 1, respectively. Let T : D0 → H 1 ,
S : D1 → H 2
be closed linear operators such that ST = 0. Let L : H 1 → H 1 be a linear bijection satisfying (Lx, x)1 ≥ 0
(x ∈ H 1 ).
(2.17)
Then we have the following lemma. Lemma 2.30
Suppose |(Lv, v)1 | ≤ kT ∗vk20 + kSvk22 ,
for every v ∈ DT ∗ ∩ DS . Then for g ∈ Ker S, there exists u ∈ DT with the following properties: T u = g, Proof.
kuk20 ≤ |(L−1 g, g)1 |.
It follows from (2.17) that (L(x + y), x + y)1 = (x + y, L(x + y))1 , (L(x + iy), x + iy)1 = (x + iy, L(x + iy))1 ,
which implies that (Lx, y)1 + (Ly, x)1 = (x, Ly)1 + (y, Lx)1 , −(Lx, y)1 + (Ly, x)1 = −(x, Ly)1 + (y, Lx)1 .
The ∂¯ Problem in Pseudoconvex Domains
101
Thus we obtain (Lx, y)1 = (x, Ly)1
(x, y ∈ H 1 ).
It follows from (2.17) that for t ∈ C we obtain (L(x + ty)1 , x + ty)1 ≥ 0. Hence for every real number t, (L(x + (Lx, y)1 ty)1 , x + (Lx, y)1 ty)1 ≥ 0, which implies that for every real number t, (Lx, x)1 + 2|(Lx, y)1 |2 t + |(Lx, y)1 |2 (Ly, y)1 t2 ≥ 0. Hence we have |(Lx, y)1 |2 ≤ (Lx, x)1 (Ly, y)1
(x, y ∈ H 1 ).
Since L is bijective, there exists g˜ ∈ H 1 such that L˜ g = g. Thus for v ∈ DT ∗ ∩ Ker S, we have |(v, g)1 |2 = |(v, L˜ g)1 |2 ≤ (Lv, v)1 (L˜ g , g˜)1 ≤ (L˜ g , g˜)1 (kT ∗ vk20 + kSvk22 ) = (L˜ g , g˜)1 kT ∗vk2 . Since (v, g)1 = 0 for v ∈ DT ∗ ∩ (Ker S)⊥ , we have for v ∈ DT ∗ , |(v, g)1 |2 ≤ (L˜ g, g˜)1 kT ∗ vk20 .
(2.18)
Hence if we define a bounded linear functional ϕ : RT ∗ → C by ϕ(T ∗ v) = (v, g)1 , then by the Hahn-Banach theorem, ϕ is extended to a bounded linear functional on H 0 . By the Riesz representaion theorem, there exists u0 ∈ H 0 such that ϕ(w) = (w, u0 )0 ,
kϕk = ku0 k0
(w ∈ H 0 ).
It follows from (2.18) that |ϕ(T ∗ v)| = |(g, v)1 | ≤
p (L˜ g , g˜)1 kT ∗vk0 ,
which implies that kϕk2 ≤ (L˜ g, g˜)1 . Consequently, ku0 k20 ≤ (L˜ g , g˜)1 .
102
Several Complex Variables and Integral Formulas
On the other hand we have ϕ(T ∗ v) = (T ∗ v, u0 )0 = (v, g)1
(v ∈ DT ∗ ).
(2.19)
Hence by (2.19) we have |(T ∗ v, u0 )0 | ≤ v1 g1 for v ∈ DT ∗ , which implies that u0 ∈ DT ∗∗ = DT . By (2.19), (v, g)1 = (v, T u0 ) for v ∈ DT ∗ , which implies that T u0 = g. Let Ω ⊂ Cn be a bounded pseudoconvex domain with C 2 boundary. Then there exist a neighborhood U of ∂Ω and a C 2 plurisubharmonic function ρ in U such that U ∩ Ω = {z ∈ U | ρ(z) < 0}. We assume that |dρ(z)| = 1 for z ∈ ∂Ω. Further, we assume that ϕ is a C 2 of Ω. For l ∈ (0, 1), define plurisubharmonic function in a neighborhood Ω ∞ χ ∈ C (R) such that (see Exercise 2.4) 2 1 (t ≤ l ) , | χ′ | ≤ χ (t) = . 0 (t ≥ 1) 1−l For 0 < ε < 12 , define
|zn |2 χε (z) = χ ε2
define Further, for f ∈ O(Ω),
gε (z) = ∂¯
.
χε (z)f (z) zn
.
We have Then gε is a ∂¯ closed C ∞ (0, 1) form on Ω. 2 |zn |2 −ϕ(z) 1 2 −ϕ(z) 2 ′ e dV (z), |gε (z)| e dV (z) = 4 |f (z)| χ ε Ωε ε2 Ω
where
Ωε = {z ∈ Ω | l ε2 ≤ |zn |2 ≤ ε2 }, and dV is the Lebesgue measure in Cn . We choose A > 1 such that Ω ⊂ Cn−1 × {zn | |zn | < A/2}. Define γε (z) =
1 , ε2 + |zn |2
ηε (z) = log(A2 γε (z)).
The ∂¯ Problem in Pseudoconvex Domains
103
Then z ∈ Ω, and for ε ∈ (0, 1/2), ηε (z) ≥ log 2. Define σ(z) =
|z|2 , log 2
ψ = ϕ + σ.
Then we have ηε (z)
n
j,k=1
|w|2 ∂2σ (z)wj w ¯k = ηε (z) ≥ |w|2 ∂zj ∂ z¯k log 2
for z ∈ Ω, w ∈ Cn , ε ∈ (0, 1/2). Consequently, ηε (z)
n
j,k=1
∂2ψ (z)wj w ¯k ≥ |w|2 ∂zj ∂ z¯k
For 0 ≤ ε < 1/2, define αε =
(z ∈ Ω, w ∈ Cn ).
1 (ε = 0) . ηε + γε (ε > 0)
We set H 0 = L2(0,0) (Ω, ψ),
H 1 = L2(0,1) (Ω, ψ),
H 2 = L2(0,2) (Ω, ψ),
and ¯ √αε u), Tε (u) = ∂(
Sε =
√ ¯ αε ∂,
T = T0 ,
S = S0 .
Then we have DTε = DT ,
DSε = DS ,
DTε∗ = DT ∗ .
Now we define a linear operator Lε : H 1 → H 1 by n−1 n−1 ε2 vj d¯ zj + 2 vj d¯ zj + vn d¯ zj = Lε vn d¯ zn . (ε + |zn |2 )2 j=1 j=1 Then Lε : H 1 → H 1 is bijective and satisfies
(Lε (x), x)1 ≥ 0, for every x ∈ H 1 .
(2.20)
Several Complex Variables and Integral Formulas
104
Lemma 2.31
Let v =
n j=1
if n
vj (z)
j=1
Proof.
Suppose v =
2 Then v ∈ DT ∗ if and only vj d¯ zj ∈ C(0,1) (Ω). ε
n j=1
∂ρ (z) = 0 ∂zj
(z ∈ ∂Ω).
2 ∩ DT ∗ . Then vj d¯ zj ∈ C(0,1) (Ω) ε
(u, T ∗v)0 = (T u, v)1
(u ∈ DT ),
which means that T ∗v = −
n
eψ
j=1
∂ (vj e−ψ ). ∂zj
We set v˜(z) =
n
vj (z)
j=1
∂ρ (z). ∂zj
Suppose there exists z 0 ∈ ∂Ω such that v˜(z 0 ) = 0. We may assume that Re v˜ > 0 in some neighborhood W of z 0 . We choose a function u ˜ ∈ Cc∞ (Cn ) u) ⊂ W . Since u˜ ∈ DT , it with the properties that u ˜ ≥ 0, u˜(z 0 ) > 0, supp (˜ follows from Green’s theorem (Theorem 2.1) that (˜ u, T ∗ v)1 = (T u ˜, v)2 = =−
Ω
u ˜
n
n ∂u ˜ vj e−ψ dV ¯j Ω j=1 ∂ z e
j=1
= (˜ u, T ∗ v)1 +
ψ ∂(vj e
−ψ
∂ z¯j
)
e
−ψ
n ∂ρ dV + u ˜vj e−ψ dS ∂ ¯j ∂Ω j=1 z
u ˜v˜e−ψ dS,
u˜v˜e−ψ dS = 0.
∂Ω
which implies that
∂Ω
This contradicts the choice of v˜ and u˜. Thus we have v˜|∂Ω = 0. Similarly we can prove the sufficiency.
The ∂¯ Problem in Pseudoconvex Domains
105
For u ∈ DT ∗ and v ∈ DT , we have ¯ √αε v), u)1 = (v, √αε T ∗ u)0 , (v, Tε∗ u)0 = (Tε v, u)1 = (∂( n
which implies that Tε∗ u =
√ 2 ∩ (Ω) αε T ∗ u. Hence, for u = uk d¯ zk ∈ C(0,1) k=1
DTε∗ ,
n
∂ √ Tε∗ u = − αε eψ (uj e−ψ ). ∂z j j=1 Theorem 2.18
2 ∩ DT ∗ , we have For 0 < ε < 1/2 and u ∈ C(0,1) (Ω) ε
(Lε u, u) ≤ Tε∗ u20 + Sε u22 .
Proof.
Using Green’s theorem (Theorem 2.1), we have Tε∗u20 + Sε u22
= (αε T ∗ u, T ∗ u)0 + (αε Su, Su)2 ¯ ε T ∗ u), u)1 = (γε T ∗ u, T ∗ u)0 + (γε Su, Su)2 + (∂(η ∂uk ∂uj ∂uk ∂uj e−ψ dV − − + ηε ∂ z¯j ∂ z¯k ∂ z¯j ∂ z¯k Ω j 0, such that " & A2 |zn | |αε | ≤ |zn |2 log + 1 ≤ B. ε2 + |zn |2 It follows from Corollary 2.5 that 1/2 ≤ |Fε |2 e−ψ dV
ˆε Ω
Ω
+
2B (1 − l )ε3
Ωε
2
dV
12
12
.
2 −ψ
|χε | |f | e
(ε2 + |zn |2 )2 |f |2 e−ψ dV
(2.23)
The first term in the right side of (2.23) converges to 0 as ε → 0. In order to investigate the second term in the right side of (2.23), we need the following lemma. Lemma 2.32 lim ε→0+
Ωε
For ϕ ∈ C ∞ (Ω), we have ϕ(z) dV (z) = (1 − l )π (|zn |2 + ε)2
ϕ(z)dVn−1 (z),
{zn =0}∩Ω
where dV and dVn−1 are the Lebesgue measures in Cn and Cn−1 , respectively. Proof. Let 0 < ε ≤ 1/2. If we choose ε sufficiently small, then there exist a constant α > 0 and compact sets E (ε) , F (ε) ⊂ Cn−1 with the following properties: √ √ E (ε) × { lε ≤ |zn | ≤ ε} ⊂ Ωε ⊂ F (ε) × { lε ≤ |zn | ≤ ε} (2.24) and µ(F (ε) − E (ε) ) ≤ αε,
(2.25)
Several Complex Variables and Integral Formulas
112
where µ is the Lebesgue measure in Cn−1 . We set z ′ = (z1 , · · · , zn−1 ), z = (z ′ , zn ). We define τ by τ (z) = ϕ(z) − ϕ(z ′ , 0). Then there exists a constant C > 0 such that |τ (z)| ≤ C|zn |. On the other hand we have ε dxn dyn rdr = 2π √ √ 2 2 2 2 lε≤|zn |≤ε (|zn | + ε) l ε (r + ε) =
(1 − l )π → (1 − l )π, (l ε + 1)(ε + 1)
as ε → 0. Hence we obtain ϕ(z) ϕ(z ′ , 0) lim dV (z) = lim dV (z) ε→0+ Ω (|zn |2 + ε)2 ε→0+ Ω (|zn |2 + ε)2 ε ε = lim (1 − l )π ϕ(z ′ , 0)dVn−1 (z ′ ) ε→0+ E (ε) = (1 − l )π ϕ(z ′ , 0)dVn−1 (z ′ ), {zn =0}∩Ω
which completes the proof of Lemma 2.32.
Since ε2 ≥ (ε2 + |zn |2 )/2 and ε ≥ (ε + |zn |2 )/2 in Dε , it follows from Lemma 2.32 that 1 (ε2 + |zn |2 )2 |f |2 e−ψ dV ε6 Ωε |f |2 e−ψ dV ≤ 16 2 2 Ωε (ε + |zn | ) ′ → 16(1 − l )π |f (z ′ , 0)|2 e−ψ(z ,0) dVn−1 H∩Ω ′ ≤ 16(1 − l )π sup e−σ(z) |f (z ′ , 0)|2 e−ψ(z ,0) dVn−1 . z∈H∩Ω
H∩Ω
Consequently, lim sup |Fε |2 e−ψ dV ≤ C ε→0
Ω
H∩Ω
′
|f (z ′ , 0)|2 e−ψ(z ,0) dVn−1 ,
(2.26)
where C = (64B 2 π)/(1 − l ) supz∈H∩Ω e−σ(z) . The following lemma is well known. So we omit the proof (See Excercise 2.3). Lemma 2.33 (Montel’s theorem) Let {uk } be a sequence of holomorphic functions in Ω which are uniformly bounded on every compact subset
The ∂¯ Problem in Pseudoconvex Domains
113
of Ω. Then there exists a subsequence {ukj } of {uk } which converges uniformly on every compact subset of Ω. Lemma 2.34 Let Ω be a bounded pseudoconvex domain in Cn with C 2 boundary whose defining function ρ satisfies |dρ| = 1 on ∂Ω. Then there exists a constant C > 0 such that for every holomorphic function f in H ∩Ω, there exists a holomorphic function F in Ω which satisfies F |H∩Ω = f and ′ |f (z ′ , 0)|2 e−ψ(z ,0) dVn−1 . |F |2 e−ψ dV ≤ C H∩Ω
Ω
Proof.
Lemma 2.34 follows from Lemma 2.33 and (2.26).
In order to prove the Ohsawa-Takegoshi extension theorem we need the following lemma. Lemma 2.35 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C 3 boundary. Then there exist a neighborhood U of ∂Ω and a C 2 strictly plurisubharmonic function ρ˜ in U such that U ∩ Ω = {z ∈ U | ρ˜(z) < 0},
|d˜ ρ(z)| = 1 (z ∈ ∂Ω).
Proof. By the definition of the strictly pseudoconvex domain, there exist a neighborhood V of ∂Ω and a strictly plurisubharmonic function ρ in V such that V ∩ Ω = {z ∈ V | ρ(z) < 0},
dρ(z) = 0 (z ∈ ∂Ω).
We may assume that dρ(z) = 0 in V . If we set ρ1 (z) = ρ(z)/|dρ(z)|, then n 1 for z ∈ ∂Ω, w ∈ Cn − {0} with j=1 ∂ρ ∂zj (z)wj = 0, we have n
j,k=1
∂ 2 ρ1 (z)wj wk > 0. ∂zj ∂ z¯k
For A > 0, we set ρ˜(z) = ρ1 (z)eAρ1 (z) , where we will determine A later. Then we have |d˜ ρ| = 1 on ∂Ω. Let P ∈ ∂Ω. Then we obtain 2 n n 2 2 ∂ ρ1 ∂ ρ˜ ∂ρ1 (P )wj w ¯k = (P )wj w ¯k + (P )wj (A + A2 ). ∂zj ∂ z¯k ∂zj ∂ z¯k j=1 ∂zj j,k=1
Several Complex Variables and Integral Formulas
114
Define X = {w | |w| = 1,
n
j,k=1
∂ 2 ρ1 (P )wj w ¯k ≤ 0}. ∂zj ∂ z¯k
Then X is compact, and X ⊂ {w | |w| = 1, Hence |
n
∂ρ1 j=1 ∂zj (P )wj |
A=
n ∂ρ1 j=1
∂zj
(P )wj = 0}.
has the minimum m > 0 in X. We set
− minw∈X
n
∂ 2 ρ1 ¯k j,k=1 ∂zj ∂ z¯k (P )wj w 2 m
+ 1.
Then for w ∈ X, n
j,k=1
∂ 2 ρ˜ (P )wj w ¯k ≥ m2 > 0. ∂zj ∂ z¯k
In case |w| = 1 and w ∈ X, we have ∂ 2 ρ1 (P )wj w ¯k > 0. ∂zj ∂ z¯k Hence for |w| = 1, we obtain n
j,k=1
∂ 2 ρ˜ (P )wj w ¯k > 0. ∂zj ∂ z¯k
(2.27)
For each P ∈ ∂Ω, there exists A = A(P ) > 0 and a neighborhood W (P ) of P such that (2.27) holds for z ∈ W (P ). Thus there exist a constant A and a neighborhood U (U ⊂ V ) of ∂Ω such that for z ∈ U and |w| = 1, n
j,k=1
∂ 2 ρ˜ (z)wj w ¯k > 0, ∂zj ∂ z¯k
which implies that ρ˜ is strictly plurisubharmonic in U .
Now we are going to prove the Ohsawa-Takegoshi extension theorem. Theorem 2.20 (Ohsawa-Takegoshi extension theorem [OHT]) Let Ω ⊂ Cn be a bounded pseudoconvex domain and let H = {z ∈ Cn | zn = 0}. Suppose ϕ is plurisubharmonic in Ω. Then there exists a constant
The ∂¯ Problem in Pseudoconvex Domains
115
C > 0 such that for every holomorphic function f in H ∩ Ω, there exists a holomorphic function F in Ω which satisfies F |H∩Ω = f and |f |2 e−ϕ dVn−1 . |F |2 e−ϕ dV ≤ C H∩Ω
Ω
Proof. We choose an increasing sequence {Ωj } of strictly pseudoconvex domains in Cn with C ∞ boundary such that Ωj are compact subsets of Ω and ∪∞ j=1 Ωj = Ω. By Lemma 2.35, we can choose the defining functions ρj for Ωj with the properties that |dρj | = 1 on ∂Ωj for j = 1, 2, · · · . Let {ϕj } be a sequence of C ∞ plurisubharmonic functions on Ωj with ϕj ↓ ϕ (Such a sequence {ϕj } exists by Theorem 1.15 and Theorem 2.16). By Theorem 2.14, we may assume that f is holomorphic in Ω. Suppose |f |2 e−ϕ dVn−1 = M < ∞. H∩Ω
It follows from Lemma 2.34 that there exist holomorphic functions Fj in Ωj such that Fj |H∩Ωj = f and ′ |f (z ′ , 0)|2 e−ϕj (z ,0) dV (z ′ ) ≤ CM. |Fj |2 e−ϕj dV ≤ C H∩Ωj
Ωj
Let K ⊂ Ω be a compact set. Then there exists a positive integer N such that K ⊂ Ωj for j ≥ N . If we set LN = minΩN e−ϕN , then it follows from Corollary 1.3 that 2 −ϕj sup |Fj | dV ≥ LN |Fj | e CM ≥ |Fj |2 dV ≥ LN C Ωj
ΩN
K
for j ≥ N . Hence {Fj } is uniformly bounded on every compact subset of Ω, and hence by the Montel theorem (Lemma 2.33), we can choose a subsequence {Fkj } of {Fj } which converges uniformly on every compact subset of Ω. We set limj→∞ Fkj = F . Then F is holomorphic in Ω and F |H∩Ω = f . Moreover we have |F |2 e−ϕ dV = lim |Fkj |2 e−ϕkj dV j→∞ K K |Fkj |2 e−ϕkj dV ≤ CM. ≤ lim j→∞
Ωkj
Berndtsson [BR2] improved Ohsawa-Takegoshi extension theorem as follows. We omit the proof.
116
Several Complex Variables and Integral Formulas
Theorem 2.21 Let Ω be a bounded pseudoconvex domain in Cn and let ϕ be plurisubharmonic in Ω. Let M = {z ∈ Ω | h(z) = 0} be a hypersurface defined by a holomorphic function bounded by 1 in Ω. Then, for any holomorphic function, f , on M there is a holomorphic function F in Ω such that F = f on M and e−ϕ 2 −ϕ dVM , |f |2 |F | e dV ≤ 4π |∂h|2 M Ω where dVM is the surface measure on M . Remark 2.2 Siu [SI2] proved that the constant C in Theorem 2.20 can $ # 1 1/2 , provided Ω ⊂ {z ∈ Cn | |zn | ≤ 1}. be chosen to be 64 9 π 1 + 4e
Exercises
2.1 2.2
2.3 2.4
Prove Theorem 2.2 for any (p, q). Show that if a family F = {fλ | λ ∈ Λ} of holomorphic functions in a domain Ω ⊂ Cn is uniformly bounded in Ω, then it is equicontinuous on every compact subset of Ω. Prove Lemma 2.33. Let a > 0 and c > 1. Construct a function f which satisfies the following conditions:
(a) f ∈ C ∞ (R), 0 ≤ f (x) ≤ 1 (x ∈ R). (b) f (x) = 1 (x ≤ 0), f (x) = 0 (x ≥ a). (c) |f ′ (x)| ≤ ac .
Chapter 3
Integral Formulas for Strictly Pseudoconvex Domains
In this chapter we study integral formulas for differential forms on bounded domains in Cn with smooth boundary. Using integral formulas, we prove H¨ older estimates for the ∂¯ problem in strictly pseudoconvex domains with smooth boundary. Moreover, by following Henkin-Leiterer [HER] and Henkin [HEN3] we prove bounded and continuous extensions of holomorphic functions from submanifolds in general position of strictly pseudoconvex domains in Cn with smooth boundary. We also study H p and C k extensions of holomorphic functions from submanifolds of strictly pseudoconvex domains in Cn with smooth boundary. Next we prove Fefferman’s mapping theorem [FEF] concerning biholomorphic mappings between strictly pseudoconvex domains with smooth boundary. The proof of Fefferman’s mapping theorem given here is based on integral formulas for strictly pseudoconvex domains obtained by Henkin-Leiterer [HER] and the method developed by Range [RAN2].
3.1
The Homotopy Formula
Let Ω ⊂ Rn be an open set and let f be a differential form with degree s on Ω. Then f is expressed by (x ∈ Ω). fi1 ···is (x)dxi1 ∧ · · · ∧ dxis f (x) = 1≤i1 n ) ∧ ω ′ (v) ∧ ω(u) − . d = < u, v >n < v, u >n < v, u >2n Moreover, we have dω ′ (v) = nω(v), and d < v, u >n ∧ ω ′ (v) ∧ ω(u) n n−1 (vj duj + uj dvj ) ∧ ω ′ (v) ∧ ω(u) = n < v, u > j=1
n
= n < v, u > ω(v) ∧ ω(u).
Several Complex Variables and Integral Formulas
120
Let ζj = xj + ixj+n for j = 1, · · · , n and let z ∈ Cn . Then
Lemma 3.3
dζ (ω ′ (ζ¯ − z¯) ∧ ω(ζ)) = n(2i)n dx1 ∧ · · · ∧ dx2n . Since
Proof.
dζ¯j ∧ dζj = (dxj − idxj+n ) ∧ (dxj + idxj+n ) = 2idxj ∧ dxj+n , we obtain n n n n dζ (−1)j+1 (ζ¯j − z¯j ) ∧ dζ¯k ∧ dζs = n ∧ dζ¯k ∧ dζs k=j
j=1
s=1
s=1
k=1
n
= n(2i)n ∧ dxj . j=1
Definition 3.4 Let A be an algebra whose elements are functions or a differential forms. Let A = (aij )ni,j=1 be an n × n matrix with entries aij ∈ A. Define detA =
σ
sgn(σ)aσ(1),1 · · · aσ(n),n ,
where σ means that the summation is performed over all permutations σ of {1, · · · , n}. Definition 3.5
Define Z = {a ∈ A | a ∧ b = b ∧ a for all b ∈ A}.
Then Z is the subalgebra of A which consists of functions and differential forms of even degree. Lemma 3.4
Suppose zi ∈ Z for i = 1, 2, · · · , n. Then we have a11 · · · z1 ∧ bk · · · z1 ∧ bs · · · a1n a21 · · · z2 ∧ bk · · · z2 ∧ bs · · · a2n . .. .. .. = 0, .. . . . a ··· z ∧ b ··· z ∧ b ··· a n1 n k n s nn
(3.1)
where bk ∈ A is contained in the k-th column, and bs is contained in the s-th column.
Integral Formulas for Strictly Pseudoconvex Domains
121
Proof. We denote by detA the determinant in (3.1). For a permutation τ = (ks), we obtain detA = sgn(σ)aσ(1)1 · · · zσ(k) bk · · · zσ(s) bs · · · aσ(n)n σ
=
σ
sgn(σ)aσ(1)1 · · · zσ(s) bk · · · zσ(k) bs · · · aσ(n)n
= sgn(τ ) =−
σ
σ
sgn(στ )aστ (1)1 · · · zστ (k) bk · · · zστ (s) bs · · · aστ (n)n
sgn(σ)aσ(1)1 · · · zσ(k) bk · · · zσ(s) bs · · · aσ(n)n = −detA,
which means that detA = 0. Lemma 3.5
For a C 1 function ψ on X, ω ′ (ψv) = ψ n ω ′ (v).
Proof.
ω ′ (v) can be written v1 1 .. ′ ω (v) = (n − 1)! . v n
dv1 · · · dv1 .. .. . . . dvn · · · dvn
Since v1 , · · · , vn ∈ Z, it follows from Lemma 3.4 that ω ′ (ψv) = det(ψv, d(ψv), · · · , d(ψv))
= det(ψv, ψdv + vdψ, · · · , ψdv + vdψ)
= det(ψv, ψdv, · · · , ψdv)
= ψ n det(v, dv, · · · , dv) = ψ n ω ′ (v).
Let Ω ⊂ Cn be a bounded domain and let f be a bounded (0, 1) form in Ω. For z ∈ Ω, define (BΩ f )(z) :=
(n − 1)! (2πi)n
ζ∈Ω
f (ζ) ∧
ωζ′ (ζ¯ − z¯) ∧ ω(ζ) , |ζ − z|2n
(3.2)
where ω(ζ) := dζ1 ∧ · · · ∧ dζn ,
ωζ′ (ζ¯ −
z¯) :=
n j=1
(−1)j+1 (ζ¯j − z¯j ) ∧ dζ¯k . k=j
122
Several Complex Variables and Integral Formulas
Let Ω ⊂ Cn be a bounded domain with C 1 boundary, and let f be a bounded function on ∂Ω. For z ∈ Ω, define (n − 1)! (B∂Ω f )(z) := (2πi)n
f (ζ)
ζ∈∂Ω
ωζ′ (ζ¯ − z¯) ∧ ω(ζ) . |ζ − z|2n
(3.3)
For (z, ζ) ∈ Cn × Cn , define n
ζ¯j − z¯j ωz,ζ (ζ¯ − z¯) ∧ ω(ζ) (−1)j+1 := ∧ (dζ¯k − d¯ zk ) ∧ ω(ζ). 2n 2n k=j |ζ − z| |ζ − z| j=1 Let f be a bounded differential form in Ω. Define (BΩ f )(z) :=
(n − 1)! (2πi)n
ζ∈Ω
f (ζ) ∧
′ ωz,ζ (ζ¯ − z¯) ∧ ω(ζ) . |ζ − z|2n
(3.4)
If f is a (0, 1) form, then (3.4) coincides with (3.2). Further, if f is a ′ function, then f (ζ)ωz,ζ (ζ¯ − z¯) ∧ ω(ζ) is at most of degree 2n − 1 with respect to ζ, and hence BΩ f = 0. Let Ω ⊂ Cn be a bounded domain with C 1 boundary, and let f be a bounded differential form on ∂Ω. Define ′ ωz,ζ (ζ¯ − z¯) ∧ ω(ζ) (n − 1)! . (3.5) f (ζ) (B∂Ω f )(z) := n (2πi) |ζ − z|2n ζ∈∂Ω If f is a function, then (3.5) coincides with (3.3). Definition 3.6 Let Ω ⊂ Rn be an open set and let α = (α1 , · · · , αn ) be a multi-index, where each αj is a nonnegative integer. Define |α| = α1 + · · · + αn , ∂α =
1 ∂xα 1
∂ |α| . n · · · ∂xα n
For f ∈ C k (Ω), define the C k -norm |f |k,Ω of f in Ω by sup |∂ α f (x)|. |f |k,Ω = |α|≤k
x∈Ω
|f |0,Ω is denoted by |f |Ω . Then {f ∈ C k (Ω) | |f |k,Ω < ∞} is a Banach space with respect to this norm. For f ∈ C(Ω) and 0 < α < 1, define the
Integral Formulas for Strictly Pseudoconvex Domains
123
α-Lipschitz norm (or the α-H¨older norm) |f |α,Ω by |f |α,Ω = |f |0,Ω +
sup z,ζ∈Ω,z=ζ
|f (z) − f (ζ)| . |ζ − z|α
We set Λα (Ω) = {f ∈ C(Ω) | |f |α,Ω < ∞}. We call Λα (Ω) the Lipschitz space (or the H¨older space) of order α. A function f ∈ Λα (Ω) is bounded and uniformly continuous in Ω. Lemma 3.6
Let Ω ⊂ Rn be an open set. Then
(a) For 0 < α < 1, Λα (Ω) is a Banach space. (b) If f ∈ Λα (Ω) then f is continuous on Ω. Proof. (a) Choose a sequence {fn } such that fn ∈ Λα (Ω), |fn −fm |α,Ω → 0. Since sup |fn (x) − fm (x)| → 0
x∈Ω
(n, m → ∞),
there exists f ∈ C(Ω) such that limn→∞ fn (x) = f (x). On the other hand, for ε > 0, there exists a positive integer N such that if n, m ≥ N , then |fn (x) − fm (x)|
0 such that for every bounded function f in Ω, BΩ f α,Ω ≤ Cα f 0,Ω . Proof. By the definition of BΩ f , there exists a constant C1 > 0 such that for z, ξ ∈ Ω n ζ¯j − z¯j ζ¯j − ξ¯j BΩ f (z) − BΩ f (ξ) ≤ C1 |f |0,Ω |ζ − z|2n − |ζ − ξ|2n dV (ζ). j=1 ζ∈Ω Thus it is sufficient to show that there exists a constant C2 > 0 and R > 0 such that for t, s ∈ Rn with |t|, |s| ≤ R, x1 − t1 x1 − s1 dV (x) ≤ C2 |t − s|| log |t − s||. − n |x − s|n x∈Rn ,|x| 0 such that x1 − s1 dV (x) ≤ C3 |t − s|1−n dV (x). n |x−t|≤|t−s|/2 |x−t|≤|t−s|/2 |x − s|
Using the polar coordinate system, there exists a constant C4 > 0 such that x1 − t1 x1 − s1 dVn (x) ≤ C4 |t − s|. − n |x − s|n |x−t|≤|t−s|/2 |x − t|
Integral Formulas for Strictly Pseudoconvex Domains
125
Similarly, there exists a constant C5 > 0 such that x1 − t1 x1 − s1 − dVn (x) ≤ C5 |t − s|. n |x − s|n |x−s|≤|t−s|/2 |x − t| On the other hand, we have x1 − t1 x1 − s1 |x − t|n − |x − s|n (x1 − t1 )|x − s|n − (x1 − s1 )|x − t|n = |x − t|n |x − s|n (x1 − t1 )(|x − s|n − |x − t|n ) − (t1 − s1 )|x − t|n = |x − t|n |x − s|n n−1
|t − s| |x − s|ν |x − t|n−ν + n n |x − s| |x − t| |x − s|n ν=0 1 1 1 ≤ n|t − s| max , + |x − s|n |x − t|n |x − s|n 1 1 + ≤ 2n|t − s| . n |x − s| |x − t|n ≤ ||x − s| − |x − t||
We set A = {x ∈ Rn | |x − t| ≥
|t − s| |t − s| , |x − s| ≥ , |x| < R}. 2 2
Then there exist constants C6 , C7 and C8 such that x1 − t1 x1 − s1 dV (y) dV (x) ≤ C − 6 |x − t|n n n |x − s| A |t−s|/2≤|y|≤2R |y| 2R dr ≤ C7 |t − s| |t−s|/2 r ≤ C8 |t − s|| log |t − s||,
which completes the proof of Lemma 3.7.
Theorem 3.1 (Bochner-Martinelli formula) Let Ω ⊂ Cn be a bounded domain with C 1 boundary and let f be a continuous function on ¯ is continuous on Ω, where the differentiation means in the Ω such that ∂f sense of distributions. Then ¯ )(z) f (z) = (B∂Ω f )(z) − (BΩ ∂f
(z ∈ Ω).
Several Complex Variables and Integral Formulas
126
Proof.
For z ∈ Ω, we set ϕ(ζ) =
′ (n − 1)! ωζ (ζ¯ − z¯) ∧ ω(ζ) . (2πi)n |ζ − z|2n
By Lemma 3.2, we have dϕ(ζ) = 0 for ζ ∈ Ω − {z}. Then we have ¯ (ζ) ∧ ϕ(ζ) d(f (ζ)ϕ(ζ)) = ∂f for ζ ∈ Ω − {z}. For any sufficiently small ε > 0, we set Ωε = {ζ ∈ Ω | |ζ − z| > ε}. It follows from Stokes’ theorem that ¯ (ζ) ∧ ϕ(ζ). ∂f (3.6) f (ζ)ϕ(ζ) − f (ζ)ϕ(ζ) = Ωε
∂Ω
|ζ−z|=ε
The left side in (3.6) can be rewritten as f (ζ)ϕ(ζ) = f (z) ϕ(ζ) + |ζ−z|=ε
|ζ−z|=ε
|ζ−z|=ε
(f (ζ) − f (z))ϕ(ζ).
It follows from Stokes’ theorem that ωζ′ (ζ¯ − z¯) ∧ ω(ζ) (n − 1)! ϕ(ζ) = (2πi)n |ζ−z|=ε ε2n |ζ−z|=ε (n − 1)! 1 = d(ωζ′ (ζ¯ − z¯) ∧ ω(ζ)) (2πi)n ε2n |ζ−z| 0, we set Uε = {(ζ, z) ∈ Cn × Cn | |ζ − z| < ε}. Since supp(v) ⊂⊂ Ω, we have for any sufficiently small ε > 0, ∂(Ω × Ω\Uε ) ∩ (Cn × supp(v)) = (∂Ω × Ω ∪ ∂Uε ) ∩ (Cn × supp(v)). It follows from Stokes’ theorem that f (ζ) ∧ ϕ(z, ζ) ∧ v(z) f (ζ) ∧ ϕ(z, ζ) ∧ v(z) − ∂Ω×Ω ∂Uε ¯ (ζ) ∧ ϕ(z, ζ) ∧ v(z) ∂f = Ω×Ω\Uε ¯ −(−1)q f (ζ) ∧ ϕ(z, ζ) ∧ ∂v(z). Ω×Ω\Uε
Hence, if we prove the equality q lim f (ζ) ∧ ϕ(z, ζ) ∧ v(z) = (−1) f (z) ∧ v(z), ε→0
∂Uε
Ω
Integral Formulas for Strictly Pseudoconvex Domains
129
then we have (3.8). We define a holomorphic mapping T : Cn × Cn → Cn × Cn by T (ξ, z) = (z + ξ, z). If we set Sε = {ξ ∈ Cn | |ξ| = ε}, then T (Sε × Cn ) = ∂Uε . Since ωz,ξ (z + ξ) ∧ v(z) = ω(ξ) ∧ v(z), we obtain T ∗ (f (ζ)∧ϕ(z, ζ)∧v(z)) =
I
′ ¯ ¯ I (n − 1)! ω (ξ) ∧ ω(ξ) ∧v(z). fI (z +ξ)d(¯ z + ξ) (2πi)n |ξ|2n
Since Sε is real 2n − 1 dimensional, we have on Sε ¯ I ∧ ω ′ (ξ) ¯ ∧ ω(ξ) = d¯ ¯ ∧ ω(ξ). d(¯ z + ξ) z I ∧ ω ′ (ξ) ¯ ∧ ω(ξ) = (−1)q ω ′ (ξ) ¯ ∧ ω(ξ) ∧ d¯ zI , Further, using the equation d¯ z I ∧ ω ′ (ξ) we have f (ζ) ∧ ϕ(z, ζ) ∧ v(z) ∂Uε
= (−1)q
z∈Cn
(n − 1)! I
(2πi)n
ξ∈Sε
fI (z + ξ)
¯ ∧ ω(ξ)
ω ′ (ξ) d¯ z I ∧ v(z). |ξ|2n
By the Bochner-Martinelli formula, the term in brackets is equal to ¯ ∧ (ξ) (n − 1)! ω ′ (ξ) fI (z) + . (3.9) (f (z + ξ) − f (z)) I I n (2πi) |ξ|2n ξ∈Sε Using the same method as the proof of Theorem 3.1, (3.9) converges to fI , which completes the proof of Theorem 3.2. Definition 3.7 (Leray map) Let Ω ⊂ Cn be a bounded open set. A C 1 mapping w : Ω × ∂Ω → Cn is called a Leray map for Ω if < w(z, ζ), ζ − z >= 0 for all (z, ζ) ∈ Ω × ∂Ω. Let w(z, ζ) be a Leray map for Ω and let Ω have a C 1 boundary. Define η(z, ζ, λ) = (1 − λ)
ζ¯ − z¯ w(z, ζ) +λ , < w(z, ζ), ζ − z > |ζ − z|2
for z ∈ Ω, 1 ≤ λ ≤ 1 and ζ ∈ ∂Ω. Further, we define ωζ′ (w(z, ζ))
:=
n j=1
(−1)j+1 wj (z, ζ) ∧ ∂¯ζ wk (z, ζ) k=j
(3.10)
Several Complex Variables and Integral Formulas
130
and ′ ωζ,λ (η(z, ζ, λ)) :=
n j=1
(−1)j+1 ηj (z, ζ, λ) ∧ dζ,λ ηk (z, ζ, λ), k=j
where η(z, ζ, λ) = (η1 (z, ζ, λ), · · · , ηn (z, ζ, λ)),
dζ,λ = dζ + dλ .
For a bounded function f on ∂Ω, define ωζ′ (w(z, ζ)) ∧ ω(ζ) (n − 1)! (L∂Ω f )(z) := f (ζ) (2πi)n ζ∈∂Ω < w(z, ζ), ζ − z >n
(z ∈ Ω). (3.11)
For a bounded (0, 1) form f on ∂Ω and z ∈ Ω, define (n − 1)! ′ (R∂Ω f )(z) := f (ζ) ∧ ωζ,λ (η(z, ζ, λ)) ∧ ω(ζ). ζ∈∂Ω (2πi)n 0≤λ≤1
(3.12)
Remark 3.1 If w(z, ζ) = ζ¯− z¯, then L∂Ω = B∂Ω and η(z, ζ, λ) = w(z, ζ), which implies that dλ ηk (z, ζ, λ) = 0, and hence R∂Ω = 0. Definition 3.8 Let Ω ⊂ Cn be a bounded domain with C 1 boundary and let w(z, ζ) be a Leray map for Ω. For (z, ζ, λ) ∈ Ω × ∂Ω × [0, 1], define n
′ ωz,ζ (w(z, ζ)) :=
j=1
(−1)j+1 wj (z, ζ) ∧ ∂¯z,ζ wk (z, ζ) k=j
and ′ ωz,ζ,λ (η(z, ζ, λ)) =
n j=1
(−1)j+1 ηj (z, ζ, λ) ∧ (∂¯z,ζ + dλ )ηk (z, ζ, λ), k=j
where ∂¯z,ζ = ∂¯z + ∂¯ζ . For a bounded differential form f on ∂Ω and z ∈ Ω, define ′ (w(z, ζ)) ∧ ω(ζ) ωz,ζ (n − 1)! f (ζ) ∧ (L∂Ω f )(z) := n (2πi) < w(z, ζ), ζ − z >n ζ∈∂Ω and (n − 1)! (R∂Ω f )(z) = (2πi)n
ζ∈∂Ω
0≤λ≤1
′ f (ζ) ∧ ωz,ζ,λ (η(z, ζ, λ)) ∧ ω(ζ).
(3.13)
(3.14)
If f is a function, then (3.13) coincides with (3.11). Further, if f is a (0, 1) form, (3.14) coincides with (3.12).
Integral Formulas for Strictly Pseudoconvex Domains
Remark 3.2
131
If w(z, ζ) = ζ¯ − z¯, then L∂Ω = B∂Ω , R∂Ω = 0.
Theorem 3.3 (Leray formula) Let Ω ⊂ Cn be a bounded domain with ¯ is continuous C 1 boundary. If f is a continuous function on Ω such that ∂f on Ω, then ¯ )(z) ¯ )(z) − (BΩ ∂f f (z) = (L∂Ω f )(z) − (R∂Ω ∂f Proof.
(z ∈ Ω).
Fix z ∈ Ω. Since < η(z, ζ, λ), ζ − z >= 1
(ζ ∈ ∂Ω, λ ∈ [0, 1]),
′ (η(z, ζ, λ)) ∧ ω(ζ)] = 0. Hence we it follows from Lemma 3.2 that dζ,λ [ωζ,λ have ′ ¯ (ζ) ∧ ω ′ (η(z, ζ, λ)) ∧ ω(ζ). (η(z, ζ, λ)) ∧ ω(ζ)] = ∂f dζ,λ [f (ζ)ωζ,λ ζ,λ ′ (η(z, ζ, λ)) ∧ ω(ζ) on By applying Stokes’ theorem to the equation f (ζ)ωζ,λ ∂Ω × [0, 1], we have ′ (η(z, ζ, λ)) ∧ ω(ζ) f (ζ)ωζ,λ ∂(∂Ω×[0,1]) ¯ (ζ)ω ′ (η(z, ζ, λ)) ∧ ω(ζ). = ∂f ζ,λ ∂Ω×[0,1]
On the other hand we have ∂(∂Ω × [0, 1]) = ∂Ω × {1} − ∂Ω × {0}. Since we have equalities ′ ωζ,λ (η(z, ζ, λ)) ∧ ω(ζ)|(ζ,λ)∈∂Ω×{0} =
ωζ′ (w(z, ζ)) ∧ ω(ζ) < w(z, ζ), ζ − z >n
and ′ ωζ,λ (η(z, ζ, λ)) ∧ ω(ζ)|(ζ,λ)∈∂Ω×{1} =
ωζ′ (ζ¯ − z¯) ∧ ω(ζ) , |ζ − z|2n
¯ = L∂Ω f − B∂Ω f . Together with the Bochner-Martinelli we obtain R∂Ω ∂f formula, we have the desired formula. Corollary 3.2 Let Ω ⊂ Cn be a bounded domain with C 1 boundary and let w(z, ζ) be a Leray map for Ω. If f is continuous on Ω and holomorphic in Ω, then f (z) = (L∂Ω f )(z)
(z ∈ Ω).
(3.15)
Several Complex Variables and Integral Formulas
132
Proof.
Corollary 3.2 follows easily from Theorem 3.3.
Definition 3.9
(3.15) is called the Cauchy-Fantappi`e formula.
Now we are going to prove the Koppelman-Leray formula. The Koppelman-Leray formula is a generalization of the Leray formula to differential forms. The Koppelman formula and Koppelman-Leray formula are called the homotopy formula (see Lieb-Michel [LIM] and Range [RAN2]). Theorem 3.4 (Koppelman-Leray formula) Let Ω ⊂ Cn be a bounded domain with C 1 boundary and let w(z, ζ) be a C 2 Leray map for Ω. Suppose ¯ is also f is a continuous (0, q) (0 ≤ q ≤ n) form on Ω such that ∂f continuous on Ω. Then ¯ + ∂(R ¯ ∂Ω + BΩ )f. (−1)q f = L∂Ω f − (R∂Ω + BΩ )∂f
(3.16)
In particular, if q = 0, then by degree reasons R∂Ω f = BΩ f = 0, and hence ¯ . f = L∂Ω f − (R∂Ω + BΩ )∂f Proof. By definition, L∂Ω f and R∂Ω f are continuous in Ω. Further, by ¯ and ∂B ¯ Ω f are all continuous in Lemma 3.7 and Theorem 3.3, BΩ f , BΩ ∂f ¯ Ω. ∂R∂Ω f is also continuous in Ω since we can perform the differentiation under the integral sign. By the Koppelman formula, it is sufficient to show that ¯ ¯ ∂Ω f = B∂Ω f − L∂Ω f + R∂Ω ∂f. ∂R If q = 0, then (3.16) is the Leray formula. Let 1 ≤ q ≤ n. It is sufficient to show that ¯ ∂Ω f )(z) ∧ v(z) = (B∂Ω f )(z) ∧ v(z) − (L∂Ω f )(z) ∧ v(z) (∂R Ω
Ω
Ω
+
Ω
¯ )(z) ∧ v(z) (R∂Ω ∂f
for v ∈ D(n,n−q) (Ω). For simplicity, we set ω = ω(ζ), ω = ωz,ζ (ζ − z). Define n
θ=
(n − 1)! (−1)j+1 ηj (z, ζ, λ) ∧ dz,ζ,λ ηk (z, ζ, λ) k=j (2πi)n j=1
and n
(n − 1)! (−1)j+1 ηj (z, ζ, λ) ∧ (∂¯z,ζ + dλ )ηk (z, ζ, λ). θ˜ = k=j (2πi)n j=1
Integral Formulas for Strictly Pseudoconvex Domains
133
Since < η(z, ζ, λ), ζ −z >= 1, it follows from Lemma 3.2 that dz,ζ,λ (θ∧ ω ) = 0. Consequently, dz,ζ,λ (θ ∧ ω) ∧ v(z) = dz,ζ,λ (θ ∧ ω ) ∧ v(z) = 0.
Since ∂ζ (θ ∧ ω) = 0, we have (∂¯z,ζ + dλ + ∂z )(θ ∧ ω) ∧ v(z) = 0. Hence we obtain ˜ ∧ ω)] ∧ v(z) = 0. [(∂¯z,ζ + dλ )(θ˜ ∧ ω) + ∂z (θ˜ ∧ ω) + (∂¯z,ζ + dλ + ∂z )((θ − θ) Since the second and the third terms in the bracket contain one of dz1 , · · · , dzn , we obtain (∂¯z,ζ + dλ )(θ˜ ∧ ω) ∧ v(z) = 0. Hence we have (∂¯ζ + dλ )(θ˜ ∧ ω) ∧ v(z) = −∂¯z (θ˜ ∧ ω) ∧ v(z). Consequently, dζ,λ (f ∧ θ˜ ∧ ω) ∧ v(z) = (∂¯ζ + dλ )(f ∧ θ˜ ∧ ω) ∧ v(z) ¯ ∧ θ˜ ∧ ω − ∂¯z (f ∧ θ˜ ∧ ω)} ∧ v(z). = {∂f It follows from Stokes’ theorem that f ∧ θ˜ ∧ ω ∧ v(z) (ζ,λ)∈∂(∂Ω×[0,1])
z∈Ω
=
z∈Ω
=
Ω
(ζ,λ)∈∂Ω×[0,1]
¯ ∧ θ˜ ∧ ω − ∂¯z ∂f
¯ )(z) ∧ v(z) − (R∂Ω ∂f
Ω
(ζ,λ)∈∂Ω×[0,1]
f ∧ θ˜ ∧ ω
∧v
(∂¯z R∂Ω f )(z) ∧ v(z).
On the other hand we have ∂(∂Ω × [0, 1]) = (−1)2n−1 ∂Ω × ∂[0, 1] = −∂Ω × {1} + ∂Ω × {0}, (n − 1)! ω ′ (w(z, ζ)) ∧ ω θ˜ ∧ ω|λ=0 = (2πi)n < w(z, ζ), ζ − z >n and (n − 1)! ω ′ (ζ¯ − z¯) ∧ ω θ˜ ∧ ω|λ=1 = . (2πi)n |ζ − z|2n
Several Complex Variables and Integral Formulas
134
Consequently,
z∈Ω
=
Ω
(ζ,λ)∈∂(∂Ω×[0,1])
f ∧ θ˜ ∧ ω
(L∂Ω f )(z) ∧ v(z) −
Ω
∧ v(z)
(B∂Ω f )(z) ∧ v(z),
which completes the proof of Theorem 3.4.
Corollary 3.3 Let Ω ⊂ Cn be a bounded domain with C 1 boundary and let w(z, ζ) be a Leray map for Ω such that w(·, ζ) is holomorphic in Ω for fixed ζ. Define for 1 ≤ q ≤ n Tq = (−1)q (R∂Ω + BΩ ). ¯ is also continuous on Let f be a continuous (0, q) form on Ω such that ∂f Ω. Then ¯ ¯ q f + Tq+1 ∂f. f = ∂T ¯ = 0, then u = Tq f is a continuous solution of the equation Moreover, if ∂f ¯ = f in Ω. ∂u Proof.
By definition, we have (L∂Ω f )(z) (n − 1)! = (2πi)n
f (ζ)
∂Ω
n
(−1)j+1
j=1
× ∧ ∂¯z,ζ wk (z, ζ) ∧ ω(ζ).
wj (z, ζ) < w(z, ζ), ζ − z >n
k=j
If q ≥ 1, then one of ∂¯z wk (z, ζ), k = 1, · · · , n, is contained in each term in the right side of the above equality. Hence L∂Ω f (z) = 0. By Theorem ¯ . It is trivial that R∂Ω f is C ∞ in Ω. By ¯ q f + Tq+1 ∂f 3.4, we have f = ∂T α Lemma 3.7, BΩ f is C in Ω. Next we will construct a Leray map for a strictly convex domain with C 2 boundary. Definition 3.10 Let Ω ⊂⊂ Rn be a domain such that Ω = {x | ρ(x) < 0}, where ρ is a C 2 defining function for Ω defined in a neighborhood of Ω. We say that Ω is strictly convex if n
j,k=1
∂2ρ (p)wj wk > 0 ∂xj ∂xk
Integral Formulas for Strictly Pseudoconvex Domains
135
for every p and 0 = w = (w1 , · · · , wn ) ∈ Rn satisfying p ∈ ∂Ω,
n ∂ρ (p)wj = 0. ∂x j j=1
We can prove the following lemma by using the same method as the proof of Corollary 1.4. We omit the proof. Lemma 3.8 Let Ω ⊂⊂ Rn be a strictly convex domain. Then there exists a defining function ρ for Ω such that n
j,k=1
Lemma 3.9 main.
∂2ρ (p)wj wk ≥ C|w|2 ∂xj ∂xk
(p ∈ ∂Ω, w ∈ Rn ).
(3.17)
Every strictly convex domain is a strictly pseudoconvex do-
Proof. Let Ω be a strictly convex domain. Then Ω = {z ∈ Cn | ρ(z) < 0}, where ρ satisfies (3.17). Let p ∈ ∂Ω. Using Taylor’s formula, we have for t = (t1 , · · · , tn ) ∈ Cn n n 2 ∂ρ ∂ ρ 1 ρ(p + t) = ρ(p) + 2Re (p)tj + (p)tj tk ∂ζ 2 ∂ζ ∂ζ j j k j=1 j,k=1
+
n
2
∂ ρ 2 ¯ ¯k (p)tj tk + o(|t| ). ∂ζ ∂ ζ j j,k=1
We set n 1 ∂ 2ρ Qp (t) = (p)tj tk 2 ∂ζj ∂ζk j,k=1
Lp (t) =
n
∂2ρ (p)tj t¯k . ∂ζj ∂ ζ¯k j,k=1
Then for tj = xj + ixn+j , n 1 ∂2ρ (p)xj xk = 2ReQp (t) + Lp (t). 2 ∂xj ∂xk j,k=1
Since Ω is strictly convex, we obtain 2ReQp (t) + Lp (t) > 0
(0 = t ∈ Cn ).
Several Complex Variables and Integral Formulas
136
Since Qp (it) = −Qp (t),
Lp (it) = Lp (t),
we obtain −2ReQp (t) + Lp (t) > 0
(0 = t ∈ Cn ).
Hence Lp (t) > 0 for t = 0, which means that Ω is strictly pseudoconvex. Definition 3.11 Let Ω1 , Ω2 ⊂ Cn be open sets. A holomorphic mapping f : Ω1 → Ω2 is called biholomorphic if f : Ω1 → Ω2 is bijective and f −1 : Ω2 → Ω1 is a holomorphic mapping. (It follows from Corollary 5.4 that if a holomorphic mapping f : Ω1 → Ω2 is bijective, then f : Ω1 → Ω2 is biholomorphic.) Lemma 3.10 (Narashimhan’s lemma) Suppose Ω ⊂⊂ Cn is a strictly pseudoconvex domain with C 2 boundary and p ∈ ∂Ω. Then there exist a neighborhood U of p and a biholomorphic mapping ϕ in U such that ϕ(U ∩Ω) is a strictly convex domain. Proof. There are a neighborhood W of ∂Ω and a strictly plurisubharmonic function ρ in W such that Ω∩W = {z ∈ W | ρ(z) < 0}. By Corollary 1.4, there exists a constant C > 0 such that n
j,k=1
∂2ρ (p)wj w ¯k ≥ C|w|2 ∂zj ∂ z¯k
(w ∈ Cn ).
We choose a coordinate system such that p = 0,
∂ρ ∂ρ (p), · · · , (p) = (1, 0, · · · , 0). ∂z1 ∂zn
Using Taylor’s formula, we have n n 2 ∂ρ 1 ∂ ρ ρ(w) = 2Re (p)wj + (p)wj wk ∂z 2 ∂z ∂z j j k j=1 j,k=1
+
n
j,k=1
2
∂ ρ (p)wj w ¯k + o(|w|2 ). ∂zj ∂ z¯k
Integral Formulas for Strictly Pseudoconvex Domains
137
We set w1′ = w1 +
n 1 ∂2ρ (p)wj wk 2 ∂zj ∂zk j,k=1
wj′ = wj
(j = 2, · · · , n).
Then w′ = ϕ(w) is a biholomorphic mapping in a neighborhood of 0 (see Corollary 5.3). Further, we have ρ ◦ ϕ−1 (w′ ) = 2Rew1′ +
n 1 ∂ 2 ρ ◦ ϕ−1 (0)wj′ w ¯k′ + o(|w′ |2 ). 2 ∂zj ∂ z¯k j,k=1
Thus there exists a neighborhood U of p such that ρ ◦ ϕ−1 is strictly convex in U . Hence ϕ(U ∩ Ω) is strictly convex.
Lemma 3.11 Let Ω ⊂⊂ Rn be a strictly convex domain with a C 2 defining function ρ. Then Ω has a C 2 boundary, that is, ρ satisfies dρ(x) = 0 for all x ∈ ∂Ω.
Proof. Suppose there exists a point p = (a1 , · · · , an ) ∈ ∂Ω such that dρ(p) = ρ(p) = 0. Using Taylor’s formula we obtain ρ(x) =
n ∂ 2 ρ(p) (xj − aj )(xk − ak ) + o(|x − p|2 ) ∂xj ∂xk
j,k=1
≥ α|x − p|2 − o(|x − p|2 ).
Hence there exists r > 0 such that ρ(x) > 0 for 0 < |x − p| < r, which contradicts that p is a boundary point of Ω. Lemma 3.12 Suppose Ω ⊂⊂ Rn is a strictly convex domain with C 2 boundary. Then Ω is geometrically convex, that is, if P1 , P2 ∈ Ω and 0 ≤ λ ≤ 1, then λP1 + (1 − λ)P2 ∈ Ω. Moreover, if P1 , P2 ∈ Ω , P1 = P2 and 0 < λ < 1, then λP1 + (1 − λ)P2 ∈ Ω. Proof. Suppose Lemma 3.12 does not hold. Then there exist λ0 with 0 < λ0 < 1 and P1 , P2 ∈ Ω such that λP1 + (1 − λ)P2 ∈ Ω for 0 ≤ λ ≤ 1, P = λ0 P1 + (1 − λ0 )P2 ∈ ∂Ω and P1 − P2 is contained in the tangent space to ∂Ω at P . We set ϕ(λ) = ρ(λP1 + (1 − λ)P2 ). Then we have ϕ(λ0 ) = ρ(P ) = 0, ϕ′ (λ0 ) = 0, We set P1 = (a1 , · · · , an ) and P2 = (b1 , · · · , bn ). By Taylor’s formula, ϕ(λ) =
n
j,k=1
∂2ρ (P )(aj − bj )(ak − bk )(λ − λ0 )2 + o(|λ − λ0 |2 ) ∂xj ∂xk
Several Complex Variables and Integral Formulas
138
≥ α|P1 − P2 |2 (λ − λ0 )2 + o(|λ − λ0 |2 ). Thus there exists r > 0 such that if 0 < |λ − λ0 | ≤ r, then ϕ(λ) > 0, which is a contradiction. One can prove the latter half similarly. Lemma 3.13 Let Ω ⊂⊂ Rn be a domain with C 2 boundary and let ρ be a defining function for Ω. Then Ω is geometrically convex if and only if n
j,k=1
∂2ρ (P )wj wk ≥ 0 ∂xj ∂xk
for all P ∈ ∂Ω and w = (w1 , · · · , wn ) ∈ Rn with
(3.18)
n
∂ρ j=1 ∂xj (P )wj
= 0.
Proof. Let Ω be geometrically convex. Suppose (3.18) does not hold. Then there exist points P ∈ ∂Ω and w ∈ Rn such that n ∂ρ (P )wj = 0, ∂xj j=1
and n
j,k=1
∂2ρ (P )wj wk = −2k < 0. ∂xj ∂xk
We may assume that P = 0, grad ρ(P ) = (0, · · · , 0.1). We set Q = tw + ε(0, · · · , 0, 1), where ε > 0 and t ∈ R will be determined later. Using Taylor’s formula we have ρ(Q) = ρ(0) +
n n ∂ρ 1 ∂2ρ (0)Qj + (0)Qj Qk + o(|Q|2 ) ∂x 2 ∂x ∂x j j k j=1 j,k=1
2
= ε+
t 2
n n
j,k=1 j,k=1
2
∂ ρ (0)wj wk + O(ε2 ) + O(εt) + o(t2 ; ε2 ) ∂xj ∂xk
= ε − kt2 + R(ε, t), where R(ε, t) satisfies |R(ε, t)| ≤ ct2 + Cε2 , and we can make c sufficiently 2 small if C is sufficiently large. We choose ε so small & that ε >> ε . If t = 0, then ρ(Q)& > 0. On the other hand, if |t| > 2ε/k, then ρ(Q) < 0. For t with |t| > 2ε/k, we set Q1 = tw + ε(0, · · · , 0, 1) and Q2 = −tw + ε(0, · · · , 0, 1). Then Q1 , Q2 ∈ Ω and (Q1 +Q2 )/2 = ε(0, · · · , 0, 1) ∈ Ω. This contradicts that Ω is geometrically convex. Conversely, assume that (3.18) holds. Suppose 0 ∈ Ω. For ε > 0 and a positive integer M , we set ρε (x) = ρ(x) + ε|x|2M /M , Ωε = {x | ρε (x) < 0}.
Integral Formulas for Strictly Pseudoconvex Domains
139
Then Ω = ∪ε>0 Ωε . If ε > 0 is sufficiently small, then Ωε is strictly convex, and hence geometrically convex. Hence Ω is geometrically convex. Lemma 3.14 Let Ω ⊂⊂ Cn be a strictly convex domain. Then there exist a neighborhood U of ∂Ω and constants ε > 0 and β > 0 such that for ζ ∈ U and z ∈ Cn with |ζ − z| ≤ ε, we have 2Re < ∂ρ(ζ), ζ − z >≥ ρ(ζ) − ρ(z) + β|ζ − z|2 , where ∂ρ(ζ) = Proof.
∂ρ(ζ) ∂ρ(ζ) ,··· , ∂ζ1 ∂ζn
.
For ζj = ξj + iξj+n and zj = xj + ixj+n , 2Re < ∂ρ(ζ), ζ − z >=
2n ∂ρ (ζ)(ξj − xj ). ∂xj j=1
Hence by Taylor’s formula, we have ρ(z) = ρ(ζ) − 2Re < ∂ρ(ζ), ζ − z >
2n 1 ∂2ρ (ζ)(ξj − xj )(ξk − xk ) + o(|ζ − z|2 ). + 2 ∂xj ∂xk j,k=1
Consequently, if we choose ε > 0 sufficiently small, then we have for ζ ∈ ∂Ω with |z − ζ| ≤ ε 2Re < ∂ρ(ζ), ζ − z >≥ ρ(ζ) − ρ(z) +
α |ζ − z|2 . 4
Theorem 3.5 Let Ω ⊂⊂ Cn be a strictly convex domain with C 2 boundary. Then 2∂ρ(ζ) is a Leray map for Ω. Proof. that
Let z ∈ Ω and ζ ∈ ∂Ω. By Lemma 3.14, there exists ε > 0 such Re < 2∂ρ(ζ), ζ − z >≥ −ρ(z) > 0,
provided |ζ − z| ≤ ε. Let |ζ − z| > ε. If we set ε ε z. zε = 1 − ζ+ |ζ − z| |ζ − z|
Several Complex Variables and Integral Formulas
140
Then by Lemma 3.12, zε ∈ Ω. Since |ζ − zε | = ε, we have Re < 2ρ(ζ), ζ − z >= 2Re
|ζ − z| < ∂ρ(ζ), ζ − zε >> 0, ε
which implies that < 2ρ(ζ), ζ − z >= 0
((z, ζ) ∈ Ω × ∂Ω).
Hence 2∂ρ(ζ) is a Leray map for Ω.
Corollary 3.4 Let Ω ⊂⊂ Cn be a strictly convex domain with C 2 bound¯ =0 ary and let f be a continuous (0, q) form, 1 ≤ q ≤ n, on Ω such that ∂f in Ω. Let w(z, ζ) = 2∂ρ(ζ) be a Leray map, where ρ is a defining function for Ω. Then u = (−1)q (R∂Ω f + BΩ f ) ¯ = f. is a continuous solution of the equation ∂u Proof.
Corollary 3.4 follows from Theorem 3.5 and Corollary 3.3.
∞
Let Ω ⊂⊂ Ω be a strictly pseudoconvex domain with C boundary. In 3.2, we will construct a Leray map w(z, ζ) for Ω which is of class C ∞ in a neighborhood of Ω × ∂Ω depending holomorphically on z for ζ fixed. In order to show that w(z, ζ) is of class C ∞ on Ω × ∂Ω, we need the following Theorem 3.6 and Theorem 3.7. We begin with Lemma 3.15. Lemma 3.15
Let Ω ⊂ Cn be a pseudoconvex open set. Define Mk = {z ∈ Cn | z1 = · · · = zk = 0}
for 1 ≤ k ≤ n. If f is holomorphic in Ω, and f (z) = 0 for z ∈ Mk ∩ Ω, then there exist holomorphic functions f1 , · · · fk in Ω such that f (z) =
k
zj fj (z)
j=1
(z ∈ Ω).
Proof. We prove the lemma by induction on k. When k = 1, we set f1 (z) = f (z)/z1 . Assume that Lemma 3.15 has already been proved for k − 1. Suppose f is holomorphic in Ω and f (z) = 0 for z ∈ Mk ∩ Ω. Since Ω∩M1 is a pseudoconvex open set in M1 , by the inductive hypothesis there exist holomorphic functions f˜2 , · · · , f˜k in Ω ∩ M1 such that f (z) =
k j=2
zj f˜j (z2 , · · · , zn )
(z ∈ Ω ∩ M1 ).
Integral Formulas for Strictly Pseudoconvex Domains
141
By Theorem 2.14, there exist holomorphic functions fj (j = 2, · · · , k) in Ω such that fj (z) = f˜j (z2 , · · · , zn ) in Ω ∩ M1 . If we set k 1 zj fj (z)) f1 (z) := (f (z) − z1 j=2
(z ∈ Ω),
then f1 is holomorphic in Ω, which completes the proof of Lemma 3.15. Lemma 3.16 Let Ω ⊂ Cn be a pseudoconvex open set and let f be a holomorphic function in Ω. Then there exist holomorphic functions f1 , · · · , fn in Ω × Ω such that n
f (w) − f (z) =
j=1
(wj − zj )fj (w, z)
for (w, z) ∈ Ω × Ω. Proof. We set g(w, z) = f (w) − f (z). Then g is holomorphic in Ω × Ω. ∗ By a change of variables zi∗ = wi − zi , zn+i = zi for i = 1, · · · , n, we have ∗ ) g(w, z) = 0 in Mn = {(w, z) ∈ Ω × Ω | w = z}. We set z ∗ = (z1∗ , · · · , z2n ∗ and h(z ) = g(w, z). By Lemma 3.15, there exist holomorphic functions h1 , · · · , hn in Ω × Ω such that ∗
h(z ) =
n
zj∗ hj (z ∗ ).
j=1
We set fj (w, z) = hj (w1 − z1 , · · · , wn − zn , z1 , · · · , zn ). Then fj , j = 1, · · · , n, are holomorphic in Ω × Ω and satisfy the equality f (w) − f (z) =
n j=1
(wj − zj )fj (w, z).
The following two theorems were proved by Range [RAN2] and used to show the smoothness of the Leray map for strictly pseudoconvex domains. Theorem 3.6 Let Ω ⊂⊂ Cn be a pseudoconvex domain and let K ⊂⊂ Ω be compact. Then there exist neighborhoods V0 and V of K, and a C ∞ function Φ(z, ζ) in V0 × ∂V with the following properties: (a) (b) (c) (d)
V0 ⊂⊂ V ⊂⊂ Ω. V has C ∞ boundary. Φ(z, ζ) is holomorphic with respect to z for fixed ζ. Φ(z, ζ) = 0 for every (z, ζ) ∈ V0 × ∂V .
Several Complex Variables and Integral Formulas
142
(e) There exist C ∞ functions wj (z, ζ), j = 1, · · · , n, in V0 × ∂V such that wj (z, ζ) are holomorphic with respect to z and satisfy Φ(z, ζ) =
n j=1
wj (z, ζ)(ζj − zj ).
O . Let ω ⊂⊂ Ω be a neighborhood Proof. We may assume that K = K Ω of K. By Theorem 1.18, there exist hk ∈ O(Ω), 1 ≤ k ≤ N , such that if A = {z ∈ ω | |hk (z)| < 1, k = 1, · · · , N },
then K ⊂ A ⊂⊂ ω. Let ∆N be the unit polydisc in CN and H = (h1 , · · · , hN ) : Ω → ∆N . Since H(K) is a compact subset of ∆N , there exists a convex neighborhood U of H(K) such that ∂U is C ∞ boundary of U and U ⊂⊂ ∆N . Let U = {t ∈ ∆N | ρ(t) < 0}. Then we have < ∂ρ(η), η − t >= 0
((t, η) ∈ U × ∂U ).
If we define Φ : A × A → C by Φ(z, ζ) =
N ∂ρ (H(ζ))(hk (ζ) − hk (z)), ∂ηk k=1
−1
then ζ ∈ (H|A ) (∂U ) and Φ(z, ζ) = 0 for z ∈ K. By the continuity, there exists a neighborhood V of K such that V0 ⊂⊂ V ⊂⊂ A, V has a C ∞ boundary and ((z, ζ) ∈ V0 × ∂V ).
Φ(z, ζ) = 0
By Lemma 3.16, there exist Qj,k ∈ O(G × G) such that hk (ζ) − hk (z) =
n j=1
Qjk (z, ζ)(ζj − zj ).
We set wj (z, ζ) =
N ∂ρ (H(ζ))Qjk (z, ζ). ∂ηk k=1
for j = 1, · · · , n. Then we obtain Φ(z, ζ) =
n j=1
wj (z, ζ)(ζj − zj ),
which completes the proof of Theorem 3.6.
Integral Formulas for Strictly Pseudoconvex Domains
143
Theorem 3.7 Let Ω ⊂⊂ Cn be a pseudoconvex domain and let K be a compact subset of Ω. Then there exist neighborhoods V0 and V of K with V0 ⊂⊂ V ⊂⊂ Ω, and a continuous linear operator Tq : C(0,q) (V ) → C(0,q−1) (V0 ) (1 ≤ q ≤ n) with the following properties: k k (V0 ) if f ∈ C(0,q) (V ). (a) for k = 0, 1, 2, · · · , Tq f ∈ C(0,q−1) ¯ = 0 on V , then ∂T ¯ q f = f on V0 . (b) If ∂f
Proof.
k For f ∈ C(0,q) (V ) and z ∈ V0 , we set
(R∂V f )(z) =
(n − 1)! (2πi)n
ζ∈∂V
0≤λ≤1
′ f (ζ) ∧ ωz,ζ,λ (η(z, ζ, λ)) ∧ ω(ζ),
where η(z, ζ, λ) is defined by using w(z, ζ) and Φ(z, ζ) in Theorem 3.6. Define Tq = (−1)q (R∂V + BV ). Then it follows from theorem 3.4 that ¯ q f )(z) + (Tq+1 ∂f ¯ )(z) f (z) = (∂T
(z ∈ V0 ),
which completes the proof of Theorem 3.7.
3.2
H¨ older Estimates for the ∂¯ Problem
Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary. Then there exist a neighborhood U of ∂Ω and a strictly plurisubharmonic C ∞ function ρ in U such that U ∩ Ω = {z ∈ U | ρ(z) < 0},
dρ(z) = 0 (z ∈ ∂Ω).
Results in 3.2, 3.3 and 3.4 are still valid under the assumption that the boundary ∂Ω is of class C 2 . However, because of Fefferman’s mapping theorem, we assume that the boundary ∂Ω is of class C ∞ . Definition 3.12 F (z, ζ) by
For ζ ∈ U and z ∈ Cn , we define the Levi polynomial
n n ∂2ρ ∂ρ (ζ)(ζj − zj ) − (ζ)(ζj − zj )(ζk − zk ). F (z, ζ) = 2 ∂ζj ∂ζj ∂ζk j=1 j,k=1
144
Several Complex Variables and Integral Formulas
F (z, ζ) is a C ∞ function in Cn × U and holomorphic with respect to z. By Taylor’s formula, there exist a constant β > 0 and ε > 0 such that for ζ ∈ U with |z − ζ| ≤ 2ε, Re F (z, ζ) ≥ ρ(ζ) − ρ(z) + β|ζ − z|2 .
(3.19)
If we choose ε > 0 sufficiently small, then for ζ ∈ ∂Ω, we have {z ∈ Cn | |ζ − z| ≤ 3ε} ⊂ U. If ε ≤ |ζ − z| ≤ 2ε, then by (3.19) we have Re F (z, ζ) ≥ ρ(ζ) − ρ(z) + βε2
(ζ, z ∈ U ).
We choose a neighborhood U1 ⊂ U of ∂Ω such that |ρ(ζ)| ≤ βε2 /3 for ζ ∈ U1 , and {z | |z − ζ| ≤ 2ε} ⊂ U for ζ ∈ U1 . We set VΩ = Ω ∪ U1 . Then for (z, ζ) ∈ VΩ × U1 with |z − ζ| ≤ 2ε, we have z, ζ ∈ U , and Re F (z, ζ) ≥
βε2 . 3
Hence we can define log F (z, ζ) for ε ≤ |ζ − z| ≤ 2ε, (z, ζ) ∈ VΩ × U1 . Choose a function χ ∈ C ∞ (Cn × Cn ) with the properties that 0 ≤ χ ≤ 1, and 1 (|ζ − z| ≤ 5ε/4) . χ(z, ζ) = 0 (|ζ − z| ≥ 7ε/4) For (z, ζ) ∈ VΩ × U1 , define ∂¯z [χ(ζ − z) log F (z, ζ)] (ε ≤ |ζ − z| ≤ 2ε) f (z, ζ) = . 0 (otherwise) ∞ (VΩ × U1 ) and ∂¯z f = 0. By Theorem 3.7, Then we have f ∈ C(0,1) there exists a neighborhood U2 of ∂Ω with U2 ⊂⊂ U1 such that if we set UΩ = Ω ∪ U2 , then UΩ ⊂⊂ VΩ . It follows from Theorem 3.7 that there ∞ exists a continuous linear operator T1 : C(0,1) (VΩ ) → C ∞ (UΩ ) such that ∂¯z T1 (f (·, ζ))(z) = f (z, ζ) for z ∈ UΩ . Define u(z, ζ) = T1 (f (·, ζ))(z). Then u ∈ C ∞ (UΩ × U2 ) and ∂¯z u = f . For (z, ζ) ∈ UΩ × U2 , we define
M (z, ζ) := e−u(z,ζ) , F (z, ζ)M (z, ζ) (|ζ − z| ≤ ε) . Φ(z, ζ) := exp[χ(ζ − z) log F (z, ζ) − u(z, ζ)] (|ζ − z| ≥ ε) Then we have the following theorem.
Integral Formulas for Strictly Pseudoconvex Domains
Theorem 3.8 (a) (b) (c) (d)
145
Φ(z, ζ) satisfies the following:
Φ(z, ζ) is a C ∞ function in UΩ × U2 . Φ(z, ζ) is holomorphic with respect to z ∈ UΩ . Φ(z, ζ) = 0 for (z, ζ) ∈ UΩ × U2 with |ζ − z| ≥ ε. There exists a C ∞ function M (z, ζ) = 0 in UΩ × U2 such that Φ(z, ζ) = F (z, ζ)M (z, ζ)
Proof.
((z, ζ) ∈ UΩ × U2 , |ζ − z| ≤ ε).
(a) holds since u(z, ζ) is C ∞ in UΩ × U2 . If |z − ζ| ≤ ε, then ∂¯z Φ(z, ζ) = F (z, ζ)e−u ∂¯z (−u) = −F (z, ζ)e−u f = 0.
If ε ≤ |z − ζ| ≤ 2ε, then ∂¯z Φ = exp[χ log F (z, ζ) − u(z, ζ)]∂¯z {χ(ζ − z) log F (z, ζ) − u(z, ζ)} = 0. If 2ε ≤ |z − ζ|, then ∂¯z Φ(z, ζ) = e−u ∂¯z (−u(z, ζ)) = −e−u f = 0, which implies that Φ(z, ζ) is holomorphic with respect to z. This proves (b). (c) and (d) follow from the definition of Φ(z, ζ). It follows from (3.19) that for (z, ζ) ∈ UΩ × U1 with ε ≤ |ζ − z| ≤ 2ε, we have Re F (z, ζ) − 2ρ(ζ) ≥ −ρ(ζ) − ρ(z) + β|ζ − z|2 ≥
βε2 . 3
Hence we can define log(F (z, ζ) − 2ρ(ζ)) for (z, ζ) ∈ UΩ × U2 with ε ≤ |ζ − z| ≤ 2ε, . For (z, ζ) ∈ UΩ × U2 , define ∂¯z [χ(ζ − z) log(F (z, ζ) − 2ρ(ζ))] (ε ≤ |ζ − z| ≤ 2ε) f˜(z, ζ) = . 0 (otherwise) Then ∂¯z f˜ = 0. It follows from Theorem 3.7 that there exists a C ∞ function ˜ = f˜. In particular, if ζ ∈ ∂Ω, then u ˜(z, ζ) in UΩ × U2 such that ∂¯z u ˜ f (z, ζ) = f (z, ζ). Hence we obtain that u ˜(z, ζ) = u(z, ζ) for ζ ∈ ∂Ω. Define '(z, ζ) = e−˜u(z,ζ) , M '(z, ζ) (F (z, ζ) − 2ρ(ζ))M (|ζ − z| ≤ ε) Φ(z, ζ) = . exp(χ log(F (z, ζ) − 2ρ(ζ)) − u ˜(z, ζ)) (|ζ − z| ≥ ε)
Then we have the following theorem which is proved in the same way as the proof of Theorem 3.8. So we omit the proof.
146
Several Complex Variables and Integral Formulas
Theorem 3.9
ζ) satisfies the following: Φ(z,
(a) (b) (c) (d)
ζ) is a C ∞ function in U × U2 . Φ(z, Ω ζ) is holomorphic with respect to z ∈ U . Φ(z, Ω ζ) = 0 for (z, ζ) ∈ U × U2 with |ζ − z| ≥ ε. Φ(z, Ω '(z, ζ) = 0 in U × U2 such that There exists a C ∞ function M Ω ζ) = (F (z, ζ) − 2ρ(ζ))M '(z, ζ) Φ(z,
((z, ζ) ∈ UΩ × U2 , |ζ − z| ≤ ε).
ζ) = Φ(z, ζ) for ζ ∈ ∂Ω. (e) Φ(z,
ζ) = 0 for (z, ζ) ∈ In particular, it follows from (c) and (d) that Φ(z, Ω × Ω\(∂Ω × ∂Ω).
Lemma 3.17 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let W = {zn = 0} ∩ UΩ . Let a function f ∈ C ∞ (W × U2 ) satisfy f (·, ζ) ∈ O(W ). Then there exist an open set V0 with Ω ⊂ V0 ⊂ UΩ and a function F ∈ C ∞ (V0 × U2 ) such that F (·, ζ) ∈ O(V0 ) (ζ ∈ U2 ), and F ((z ′ , 0), ζ) = f ((z ′ , 0), ζ)
((z ′ , 0) ∈ {zn = 0} ∩ V0 , ζ ∈ U2 ).
Proof. We define a mapping π : Cn → Cn by π(z ′ , zn ) = (z ′ , 0). Since W and UΩ − π −1 (W ) are closed disjoint subsets of UΩ , there exists a function χ ∈ C ∞ (UΩ ) with the properties that χ = 1 in an open subset of UΩ containing W , and χ = 0 in an open subset of UΩ containing UΩ − π −1 (W ). We set ∂¯z {χ(z)f (π(z), ζ)} α(z, ζ) = . zn ∞ It follows from the definition of χ that α ∈ C(0,1) (UΩ × U2 ). By Theorem 3.7, there exist an open set V , V0 (Ω ⊂ V0 ⊂⊂ V ⊂ UΩ ), and a continuous ∞ ¯ 1 (α(·, ζ))(z) = α(z, ζ) linear operator T1 : C(0,1) (V ) → C ∞ (V0 ) such that ∂T for z ∈ V0 . Define g(z, ζ) = T1 (α(·, ζ))(z). Then g ∈ C ∞ (V0 × U2 ). If we set F (z, ζ) = χ(z)f (π(z), ζ) − zn g(z, ζ), then F (·, ζ) is holomorphic in V0 . F ∈ C ∞ (V0 × U2 ) and F (π(z), ζ) = f (π(z), ζ), which completes the proof of Lemma 3.17.
The following lemma is the parametrized version of Lemma 3.15. Lemma 3.18 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary. Define Mk = {z ∈ Cn | z1 = · · · = zk = 0}
Integral Formulas for Strictly Pseudoconvex Domains
147
for 1 ≤ k ≤ n. If f (z, ζ) is of class C ∞ in UΩ × U2 and holomorphic with respect to z ∈ UΩ for ζ ∈ U2 fixed, and f (z, ζ) = 0 for z ∈ Mk ∩ UΩ , then there exist an open set V0 (Ω ⊂ V0 ⊂ UΩ ) and C ∞ functions f1 (z, ζ), · · · fk (z, ζ) in V0 × U2 which are holomorphic with respect to z ∈ Ω for ζ fixed such that f (z, ζ) =
k
zj fj (z, ζ)
((z, ζ) ∈ V0 × U2 ).
j=1
Proof. 3.15.
Lemma 3.18 follows from Lemma 3.17 and the proof of Lemma
Lemma 3.19 Let Ω ⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let f (z, ζ) be of class C ∞ in UΩ × U2 and holomorphic with respect to z ∈ UΩ for ζ ∈ U2 fixed. Then there exist an open set V0 (Ω ⊂ V0 ⊂ UΩ ) and C ∞ functions f1 , · · · , fn in V0 × V0 × U2 which are holomorphic with respect to (z, w) ∈ V0 × V0 for ζ ∈ U2 fixed such that f (w, ζ) − f (z, ζ) =
n j=1
(wj − zj )fj (w, z, ζ)
for (w, z, ζ) ∈ V0 × V0 × U2 . Proof. 3.16.
Lemma 3.19 follows from Lemma 3.18 and the proof of Lemma
Theorem 3.10 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary. Then there exist a neighborhood V0 of Ω and C ∞ functions wj (z, ζ) for j = 1, · · · , n in V0 × U2 such that Φ(z, ζ) =
n j=1
(zj − ζj )wj (z, ζ)
((z, ζ) ∈ V0 × U2 ) .
Moreover, wj (z, ζ), j = 1, · · · , n, are holomorphic with respect to z ∈ V0 for ζ fixed. Proof. By Lemma 3.19 there exist a neighborhood V0 and functions fj ∈ C ∞ (V0 × V0 × U2 ) such that fj (z, w, ζ) are holomorphic with respect to (z, w) ∈ V0 × V0 for ζ fixed and Φ(z, ζ) − Φ(w, ζ) =
n j=1
(zj − wj )fj (z, w, ζ).
Several Complex Variables and Integral Formulas
148
We set w = ζ. Because of Φ(ζ, ζ) = 0, we have Φ(z, ζ) =
n j=1
(zj − ζj )fj (z, ζ, ζ).
If we set wj (z, ζ) = fj (z, ζ, ζ), then each wj (z, ζ) satisfies the desired conditions. Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and w(z, ζ) be the Leray map for Ω. For any f ∈ O(Ω) which is continuous on Ω, the Cauchy-Fantappi´e formula (3.15) is expressed by f (z) = (L∂Ω f )(z) =
(n − 1)! (2πi)n
f (ζ)
ζ∈∂Ω
ωζ′ (w(z, ζ)) ∧ ω(ζ) . < w(z, ζ), ζ − z >n
(3.20)
Definition 3.13 Let Ω be a strictly pseudoconvex domain in Cn with C ∞ boundary. The kernel of the Cauchy-Fantappi´e formula (3.20) is called the Henkin-Ramirez kernel. We need the following lemma in order to prove 21 -H¨older estimate for the ∂¯ problem in Ω. We omit the proof (see Exercise 3.1). Lemma 3.20 Let Ω be a bounded domain in Rn with C 1 boundary. Suppose f ∈ C 1 (Ω) and that for some 0 < α < 1 there exists a constant C > 0 such that df (x) ≤ C[dist(x, ∂Ω)]α−1 ] for all x ∈ Ω. Then f ∈ Λα (Ω). Now we are going to prove the 12 -H¨older estimate for the ∂¯ problem in strictly pseudoconvex domains with smooth boundary. Theorem 3.11 Let Ω be a strictly pseudoconvex domain in Cn with C ∞ boundary. Let w be a Leray map defined in Theorem 3.10. Then for any bounded differential form f on ∂Ω R∂Ω f 1/2,Ω ≤ Cf 0,Ω . Proof.
By definition R∂Ω is expressed by
(R∂Ω )(z) =
∂Ω×[0,1]
f∧
∧dλ ∧ ω(ζ).
n−2 s=0
ps det1,1,n−s−2,s
w ζ¯ − z¯ ∂¯ζ w dζ¯ − d¯ z , , , Φ |ζ − z|2 Φ |ζ − z|2
Integral Formulas for Strictly Pseudoconvex Domains
149
Hence the coefficients of the form R∂Ω f are linear combinations of the integrals of the following type ψ E(z) = ∧ dζ¯j ∧ ω(ζ), n−s−1 |ζ − z|2s+2 j=m Φ ∂Ω where ψ satisfies |ψ| ≤ C|ζ − z|. By Lemma 3.20, it is sufficient to prove that, for j = 1, · · · , n, ∂E(z) ∂E(z) −1/2 , . ∂zj ∂ z¯j ≤ Cf 0,Ω |ρ(z)|
Therefore it is sufficient to show that for every ξ ∈ ∂Ω, there are a neighborhood U and a constant C > 0 such that dσ2n−1 ≤ C|ρ(z)|−1/2 n−s−1 |ζ − z|2s+2 |Φ(z, ζ)| ∂Ω∩U
and
∂Ω∩U
dσ2n−1 ≤ C|ρ(z)|−1/2 , |Φ(z, ζ)|n−s |ζ − z|2s+1
where dσ2n−1 is the surface measure on ∂Ω. We can choose a local coordinate system t = (t1 , · · · , t2n−1 ) in U ∩ ∂Ω such that t1 = Im Φ(z, ζ) and |t| ≈ |z − ζ|. It follows from (3.19) that dt1 · · · dt2n−1 dσ2n−1 ≤ C , n−s 2s+1 2 n−s |t|2s+1 |ζ − z| |t| 1/2. Then (u − z¯2 )/ log(z1 − 1) is holomorphic in Ω. Let ε be such that 0 < 2ε < 1. We set √ C1 = {(z1 , z2 ) ∈ C2 | z1 = 1 − ε, |z2 | = ε}, C2 = {(z1 , z2 ) ∈ C2 | z1 = 1 − 2ε, |z2 | =
√
ε}.
Then C1 , C2 ⊂ Ω. By the Cauchy integral formula we have z¯2 dz2 2πi u(1 − ε, z )dz = = . 2 2 √ log(−ε) √ log(−ε) |z2 |= ε |z2 |= ε
√ |z2 |= ε
u(1 − 2ε, z2 )dz2 =
2πi . log(−2ε)
Since u ∈ Λα (Ω), there exists a constant C > 0 such that for every 0 < 2ε < 1, we have 1 1 α−1/2 . log(−ε) − log(−2ε) ≤ Cε Consequently, for any 0 < 2ε < 1 we have
log 2 = | log(−2ε) − log(−ε)| ≤ Cεα−1/2 | log(−ε) log(−2ε)|,
Integral Formulas for Strictly Pseudoconvex Domains
which is a contradiction.
151
As an application of Theorem 3.11, we have the following lemma. Lemma 3.21 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let f be a holomorphic function in Ω that is continuous on Ω. Then f can be approximated uniformly on Ω by functions holomorphic in a neighborhood of Ω. Proof. Let U = {Ui | i = 1, · · · , N } be a finite open cover of Ω. Choose N χj ∈ Cc∞ (Uj ) such that j=1 χj = 1 on Ω. Define fj = L∂Ω (χj f ),
where w is the Leray map defined in Theorem 3.10. By Corollary 3.2 we have f=
N
fj
j=1
and each fj is holomorphic in some neighborhood of Ω\(∂Ω ∩ Uj ). By Theorem 3.3 we have ¯ j ) + BΩ (f ∂χ ¯ j ). fj = χj f + R∂Ω (f ∂χ ¯ j ) is continuous on Ω. By Theorem It follows from Lemma 3.7 that BΩ (f ∂χ 3.11 we have R∂Ω f 1/2,Ω ≤ Cf 0,Ω . w Hence R∂Ω f is continuous on Ω. Hence each fj is continuous on Ω. It is sufficient to show that each fj can be approximated uniformly on Ω by functions holomorphic in a neighborhood of Ω. The required approximation can be obtained by a shift in the direction of the normal vector of ∂Ω at some point in ∂Ω ∩ Uj .
Remark 3.3 Lemma 3.21 was first proved by Lieb [LI1]. The above proof is due to Henkin-Leiterer [HER]. ¯ = 0. We Suppose f is a continuous (0, q) form on Ω such that ∂f q set Tq = (−1) (R∂Ω + BΩ ). By (3.16) or by Corollary 3.3 we have ¯ q f , which means that Tq f is a solution of the ∂¯ problem. Using f = ∂T Tq f , Henkin [HEN2], Grauert-Lieb [GRL], Lieb [LI2; LI3], Kerzman [KER], Ovrelid [OV], Henkin-Romanov [HEV] and Krantz [KR1] obtained Lp and H¨older estimates for the ∂¯ problem in strictly pseudoconvex domains with
152
Several Complex Variables and Integral Formulas
smooth boundary. We proved 12 -H¨older estimate for the ∂¯ problem in strictly pseudoconvex domains with smooth boundary in Corollary 3.5. In 4.2 we will prove Lp estimates for the ∂¯ problem in strictly pseudoconvex domains with smooth boundary by applying the Berndtsson-Andersson formula. Bruna and Burgu´es [BRG] obtained 21 -H¨older and Lp estimates for the ∂¯ problem in strictly pseudoconvex domains with nonsmooth boundary using the Berndtsson-Andersson formula. Siu [SI1] and Lieb-Range [LIR] studied the differentiability for solutions of the ∂¯ problem in strictly pseudoconvex domains with smooth boundary. Moreover, in the finite intersection of strictly pseudoconvex domains with smooth boundary, Michel [MIC] and Michel-Perotti [MIP] obtained C k estimates, and Range-Siu [RAS] and Menini [MEN] obtained Lp and H¨ older estimates for the ∂¯ problem. Menini used the Bendtsson-Andersson formula. Range [RAN1], DiederichFornaess-Wiegerinck [DIK] and Chen-Krantz-Ma [CHK] obtained H¨ older and Lp estimates for the ∂¯ problem in real or complex ellipsoids. BrunaCastillo [BRJ], Polking [POL] and Range [RAN4] obtained H¨ older and Lp ¯ estimates for the ∂ problem in some convex domains. S.C. Chen [CH] and Z. Chen [CHE] investigated the real analyticity for solutions of the ∂¯ problem in certain convex domains. Fischer-Lieb [FIL], Ho [Ho1], Schmalz [SCH] and Ma [MA] investigated the ∂¯ problem in q-convex domains. Fleron [FLE], Ho [Ho2] and Verdera [VER] obtained H¨ older and uniform estimates for the ∂¯ problem in some domains. On the other hand, Kohn [KON] proved the global regurality for solutions of the ∂¯ problem in pseudoconvex domains in Cn with smooth boundary, that is, if Ω is a pseudoconvex domain in Cn with C ∞ boundary and f is a C ∞ (0, 1) form ¯ = 0, then there exists a C ∞ function u on Ω such that ∂u ¯ =f on Ω with ∂f (see D’Angelo [DA]). Fefferman-Kohn [FEK] studied H¨ older estimates for ¯ the ∂ problem in pseudoconvex domains of finite type in C2 with smooth boundary. Range [RAN3] also investigated the ∂¯ problem in pseudoconvex domains of finite type in C2 with smooth boundary using the homotopy formula.
3.3
Bounded and Continuous Extensions
Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with smooth boundary and let X be a submanifold in a neighborhood of Ω which intersects ∂Ω transversally. In 1972, Henkin [HEN3] proved that every bounded holomorphic function in V = X ∩Ω can be extended to a bounded holomorphic func-
Integral Formulas for Strictly Pseudoconvex Domains
153
tion in Ω. Moreover, Henkin [HEN3] proved that if f is holomorphic in V that is continuous on V , then f can be extended to a holomorphic function in Ω that is continuous on Ω. In 1984, Henkin and Leiterer [HER] extended Henkin’s results to strictly pseudoconvex domains with non-smooth boundary in a Stein manifold without assuming the transversality. In 3.3 and 3.4, we only treat the smooth domain and assume the transversality. We prove first the bounded extension from complex hypersurfaces by following the method of Henkin-Leiterer [HER], and then the continuous and bounded extensions from submanifolds by following the method of Henkin [HEN3]. Let Ω be a strictly pseudoconvex domain in Cn with C ∞ boundary. Then there exists a neighborhood U of ∂Ω and a strictly plurisubharmonic C ∞ function ρ in U such that U ∩ Ω = {z ∈ U | ρ(z) < 0},
dρ(z) 6= 0 (z ∈ ∂Ω).
Let U2 be the open set in Theorem 3.8. We choose ε0 > 0 such that {ζ ∈ U | |ρ(ζ)| < 2ε0 } ⊂⊂ U2 . Let χ ∈ D(Cn ) be a function with the following properties: (a) 0 ≤ χ ≤ 1. (b) χ(ζ) = 1 for ζ ∈ U with ρ(ζ) ≥ −ε0 . (c) χ(ζ) = 0 for ζ ∈ (Ω − U ) ∪ {ζ ∈ U | ρ(ζ) ≤ −2ε0 }. Define ωζ
χ(ζ)w(z, ζ) e ζ) Φ(z,
!
n
= ∧ dζ j=1
χ(ζ)wj (z, ζ) e ζ) Φ(z,
!
.
(3.21)
By Theorem 3.9 (d), the differential form in (3.21) is continuous with respect to (z, ζ) ∈ Ω × Ω. For an L1 function f in Ω, define ! Z χ(ζ)w(z, ζ) n! f (ζ)ωζ ∧ ω(ζ) LΩ f (z) := e ζ) (2πi)n Ω Φ(z,
Definition 3.14
(z ∈ Ω).
e ζ) are holomorphic with respect to z, LΩ f is holoSince w(z, ζ) and Φ(z, morphic in Ω. Definition 3.15
For 0 ≤ λ ≤ 1, define
η˜j (z, ζ, λ) := (1 − λ)
ζ¯j − z¯j χ(ζ)wj (z, ζ) +λ , e ζ) |ζ − z|2 Φ(z,
η˜ = (˜ η1 , · · · , η˜n ),
Several Complex Variables and Integral Formulas
154
and n ω(˜ η (z, ζ, λ)) := ∧ (∂¯z,ζ + dλ )˜ ηj (z, ζ, λ). j=1
1
Definition 3.16
For an L function f in Ω, define n! RΩ f (z) := f (ζ) ∧ ω(˜ η (z, ζ, λ)) ∧ ω(ζ) (2πi)n (ζ,λ)∈Ω×[0,1]
(z ∈ Ω).
Then RΩ f (z) is continuous in Ω. If f is a (0, q) form, then RΩ f is a (0, q−1) form. In particular, if f is a function, then RΩ f = 0. Theorem 3.12 C ∞ boundary.
Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with
¯ is an L1 form in Ω. (a) Suppose f is an L1 function in Ω such that ∂f Then ¯ (z) f (z) = LΩ f (z) + RΩ ∂f
(z ∈ Ω).
¯ is an L1 (b) Suppose f is an L1 (0, q) (1 ≤ q ≤ n) form in Ω such that ∂f form in Ω. Then ¯ Ω f (z) + RΩ ∂f ¯ (z) f (z) = ∂R
(z ∈ Ω).
Proof. Let 0 ≤ q ≤ n. Suppose f is a continuous (0, q) form on Ω such ¯ is continuous on Ω. For z ∈ Ω, ζ ∈ Ω\{z} and 0 ≤ λ ≤ 1, define that ∂f ′ (˜ η (z, ζ, λ)) := ωz,ζ,λ
n j=1
(−1)j+1 η˜j (z, ζ, λ) ∧ (∂¯z,ζ + dλ )˜ ηk (z, ζ, λ). k=j
Then ωz,ζ,λ (˜ η (z, ζ, λ)) is continuous for (z, ζ, λ) with z ∈ Ω, ζ ∈ Ω\{z}, 0 ≤ ζ) = 0 for (z, ζ) ∈ Ω × Ω, each term of ωz,ζ,λ (˜ η (z, ζ, λ)) λ ≤ 1. Since Φ(z, involving dλ is equal to O(|ζ − z|−(2n−2) ). Hence ′ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ), (z ∈ Ω) (ζ,λ)∈Ω×[0,1]
is differentiable with respect to z ∈ Ω. Differentiating under the integral sign and taking into account that dimR (Ω × [0, 1]) is odd, we have ′ ∂¯z f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ) Ω×[0,1] ′ =− ∂¯z [f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ)], Ω×[0,1]
Integral Formulas for Strictly Pseudoconvex Domains
155
and ′ (∂z,ζ + dλ )ωz,ζ,λ (˜ η (z, ζ, λ)) = nω(˜ η (z, ζ, λ)).
Consequently, ′ dζ,λ [f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ)] ′ ¯ = ∂ζ f (ζ) ∧ ω (˜ η (z, ζ, λ)) ∧ ω(ζ) z,ζ,λ
η (z, ζ, λ)) ∧ ω(ζ) +(−1)q nf (ζ) ∧ ω(˜ ′ ¯ −∂z [f (ζ) ∧ ω (˜ η (z, ζ, λ)) ∧ ω(ζ)]. z,ζ,λ
Since ∂(Ω × [0, 1]) = ∂Ω × [0, 1] + (−1)dimR Ω × ∂([0, 1]) = ∂Ω × [0, 1] − Ω × {0} + Ω × {1},
it follows from Stokes’ theorem that (2πi)n ′ RΩ f (z) ∂¯ζ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ) + (−1)q (n − 1)! Ω×[0,1] ′ (˜ η (z, ζ, λ)) ∧ ω(ζ)] − ∂¯z [f (ζ) ∧ ωz,ζ,λ Ω×[0,1]
= − +
∂Ω×[0,1]
′ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ)
Ω×{0}
′ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ)
Ω×{1}
′ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ).
If ζ ∈ ∂Ω, then χ(ζ) = 1, and hence η˜(z, ζ, λ) = η(z, ζ, λ). Thus we obtain
∂Ω×[0,1]
′ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ) =
(2πi)n R∂Ω f (z). (n − 1)!
On the other hand we have (2πi)n ′ BΩ f (z). f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ) = (n − 1)! Ω×{1}
Several Complex Variables and Integral Formulas
156
ζ) is holomorphic with respect to z, we Since η(z, ζ, 0) = χ(ζ)w(z, ζ)/Φ(z, have χ(ζ)w(z, ζ) ′ ′ ∧ ω(ζ). f (ζ) ∧ ωζ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ) = ζ) Φ(z, Ω Ω×{0} We set Tq = (−1)q (R∂Ω + BΩ ). For z ∈ Ω, we have RΩ f (z) = Tq f (z)
(n − 1)! ′ ∂¯z f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ) +(−1)q+1 (2πi)n Ω×[0,1] χ(ζ)w(z, ζ) ′ f (ζ) ∧ ωζ + ∧ ω(ζ) ζ) Φ(z, Ω
′ ∂¯ζ f (ζ) ∧ ωz,ζ,λ + (˜ η (z, ζ, λ)) ∧ ω(ζ) . Ω×[0,1]
¯ , then we have In the above equality, if we replace f by ∂f ¯ (z) + (−1)q ¯ (z) = Tq+1 ∂f RΩ ∂f
∂¯z
Ω×[0,1]
(n − 1)! × (2πi)n
′ ∂¯ζ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ)
(3.22)
for z ∈ Ω. By degree reasons we have for q ≥ 1 χ(ζ)w(z, ζ) ′ f (ζ) ∧ ωζ ∧ ω(ζ) = 0. ζ) Φ(z, Ω
(3.23)
+
Ω
∂¯ζ f (ζ) ∧ ωζ′
χ(ζ)w(z, ζ) ζ) Φ(z,
∧ ω(ζ)
ζ) are holomorphic with respect to z, we have Since w(z, ζ) and Φ(z, χ(ζ)w(z, ζ) ′ ¯ ∧ ω(ζ) = 0. (3.24) ∂z f (ζ) ∧ ωζ ζ) Φ(z, Ω
It follows from (3.22), (3.23) and (3.24) that
¯ Ω f (z) = ∂T ¯ q f (z) ∂R
Integral Formulas for Strictly Pseudoconvex Domains q+1 (n
+(−1)
− 1)! ¯ ∂z (2πi)n
Ω×[0,1]
157
′ (˜ η (z, ζ, λ)) ∧ ω(ζ). ∂¯ζ f (ζ) ∧ ωz,ζ,λ
(3.25)
First we prove (b). Suppose q ≥ 1. Since the last integral in the right side of (3.22) equals 0 by degree reasons, it follows from (3.22) and (3.25) that ¯ ¯ + ∂R ¯ Ω f = ∂T ¯ q f + Tq+1 ∂f. RΩ ∂f ¯ , which proves (b). ¯ q f + Tq+1 ∂f By Corollary 3.3, we have f = ∂T Next we prove (a). By Theorem 3.4, we have ¯ f = L∂Ω f + T1 ∂f. ζ) = Φ(z, ζ) for ζ ∈ ∂Ω, we have Since Φ(z, (n − 1)! χ(ζ)w(z, ζ) ′ L∂Ω f (z) = ∧ ω(ζ). f (ζ)ωζ ζ) (2πi)n ∂Ω Φ(z,
Since
dζ ωζ′
χ(ζ)w(z, ζ) ζ) Φ(z,
= nωζ
χ(ζ)w(z, ζ) ζ) Φ(z,
,
it follows from Stokes’ theorem that
(n − 1)! χ(ζ)w(z, ζ) ′ L∂Ω f (z) = ∧ ω(ζ) + LΩ f (z). ∂¯ζ f (ζ)ωζ ζ) (2πi)n Φ(z, Ω (3.26) Since q = 0, we obtain by degree reasons ′ ∂¯ζ f (ζ) ∧ ωz,ζ,λ (˜ η (z, ζ, λ)) ∧ ω(ζ) = 0. Ω×[0,1]
Hence we have together with (3.22) and (3.26) ¯ (z) + L∂Ω f (z) − LΩ f (z). ¯ (z) = T1 ∂f RΩ ∂f This proves (a). In the general case, Theorem 3.12 is proved using the fact ¯ can be approximated in L1 norm by continuous functions that f and ∂f with compact support. Definition 3.17 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary. Define X := {z ∈ Cn | zn = 0}.
Several Complex Variables and Integral Formulas
158
For ζ = (ζ1 , · · · , ζn ) ∈ Cn , define ζ ′ = (ζ1 , · · · , ζn−1 ). Further, we define ∂ζ ′ :=
n−1 j=1
∂ dζj , ∂ζj
dζ ′ := ∂¯ζ ′ + ∂ζ ′ ,
∂¯ζ ′ :=
n−1 j=1
∂ ¯ dζj , ∂ ζ¯j
ωζ ′ (ζ) = dζ1 ∧ · · · ∧ dζn−1 ,
and (w′ (z, ζ)) := (w1 (z, ζ), · · · , wn−1 (z, ζ)), where w(z, ζ) = (w1 (z, ζ), · · · , wn (z, ζ)) is the Leray map defined in Theorem 3.10. Define n−1 χ(ζ)w (z, ζ) χ(ζ)(w(z, ζ))′ j := ∧ ∂¯ζ ′ . ωζ ′ ζ) ζ) j=1 Φ(z, Φ(z, By Theorem 3.9, there exists a neighborhood U∂Ω\X of ∂Ω\X such that ζ) = 0 Φ(z,
(ζ ∈ X ∩ Ω, z ∈ Ω ∪ U∂Ω\X ).
We set V = X ∩ Ω. For f ∈ O(V ) ∩ L1 (V ) and z ∈ Ω ∪ U∂Ω\V , define χ(ζ)(w(z, ζ))′ (n − 1)! Ef (z) := f (ζ)ωζ ′ (3.27) ∧ ωζ ′ (ζ). ζ) (2πi)n−1 V Φ(z, The following theorem follows from Theorem 3.12.
Theorem 3.13 Let f ∈ O(V ) ∩ L1 (V ). Then Ef is holomorphic in Ω ∪ U∂Ω\V and satisfies Ef (z) = f (z)
(z ∈ V ).
Definition 3.18 Let Ω ⊂⊂ Cn be a domain with C ∞ boundary. Suppose such that of Ω and a C ∞ function ρ in Ω there exist a neighborhood Ω Ω = {z ∈ Ω | ρ(z) < 0}. Let X be a k dimensional complex submanifold in a neighborhood of Ω. We set V = X ∩ Ω. Let P ∈ ∂V . Then there exist (P ) a neighborhood U (P ) of P and a holomorphic coordinate system f1 , · · · , (P ) fn in U (P ) such that (P )
(P )
U (P ) ∩ X = {z ∈ U (P ) | f1 (z) = · · · = fn−k (z) = 0}. We say that X intersects ∂Ω transversally if (P )
(P )
df1 (P ) ∧ · · · ∧ dfn−k (P ) ∧ dρ(P ) = 0
Integral Formulas for Strictly Pseudoconvex Domains
159
for every point P ∈ ∂Ω ∩ X. Moreover, in this case we say that the submanifold X ∩ Ω of Ω is a submanifold in general position of Ω. In what follows we assume that X intersects ∂Ω transversally. Definition 3.19 Let U2 be the neighborhood of ∂Ω in Theorem 3.8. For (z, ζ) ∈ Ω × U2 , define Φ∗ (z, ζ) = Φ(ζ, z), ∗
∗
w(z, ζ) = −w(ζ, z),
w′ (z, ζ) = (∗ w1 (z, ζ), · · · , ∗ wn−1 (z, ζ)).
The following lemma was proved by Henkin-Leiterer [HER]. In their proof the transversality is not assumed. For simplicity, we assume that X intersects ∂Ω transversally in the following lemma. Lemma 3.22
If f is a bounded holomorphic function in V = X ∩Ω, then
Ef (z) (−1)n−1 = zn (2πi)n−1 for z ∈ ∂Ω\V .
X∩Ω
f (ζ)det1,n−1
∗
w(z, ζ) ¯ χ(ζ)w(z, ζ) , ∂ζ ′ ζ) Φ∗ (z, ζ) Φ(z,
∧ ωζ ′ (ζ)
Proof. By applying the expansion formula of the determinant to the n-th column, we have ∗ w ¯ χw ′ , ∂ (−1)n det1,n−1 ζ Φ∗ Φ ∗ ∗ ∗ w2 wn w∗1 · · · ∗ ∗ Φ Φ Φ ∂¯ ′ χw1 ∂¯ ′ χw2 · · · ∂¯ ′ χwn ζ ζ ζ Φ Φ Φ = (−1)n ··· ··· ··· ··· ¯ χw1 ¯ χw2 n ∂ζ ′ ∂ζ ′ · · · ∂¯ζ ′ χw Φ Φ Φ ∗ wn χw′ ¯ ′ = − ∗ detn−1 ∂ζ Φ Φ ∗ ′ χwn w ¯ χw′ ¯ ′ ′ +(n − 1)∂ζ . ∧ det1,n−2 , ∂ζ Φ∗ Φ Φ We have by the definition of the determinant ′ χwj′ n−1 χw χw′ ¯ ¯ detn−1 ∂ζ ′ = (n − 1)! ∧ ∂ζ ′ . = (n − 1)!ωζ ′ j=1 Φ Φ Φ
(3.28)
Several Complex Variables and Integral Formulas
160
For ζ ∈ V we have n ∗ wj (z, ζ)(ζj − zj ) = −1. Φ∗ (z, ζ) j=1
Consequently, −zn
n−1 ∗ wn wj (ζj − zj ) ∗ . = −1 − ∗ Φ Φ j=1
∗
(3.29)
Since n χwj (ζj − zj ) χΦ + = 0, Φ Φ j=1
it follows that for ζ ∈ V
χwn = zn ∂¯ζ ′ Φ
n−1 j=1
χwj χΦ + ∂¯ζ ′ . (ζj − zj )∂¯ζ ′ Φ Φ
Therefore, together with (3.28) and (3.29), we obtain for ζ ∈ V , ∗ w ¯ χw ′ zn (−1)n det1,n−1 , ∂ ζ Φ∗ Φ ′ n−1 ∗ χw wj = −1 − (ζj − zj ) ∗ (n − 1)!ωζ ′ Φ Φ j=1
∗ ′ χwj w ¯ χw′ ′ (ζj − zj )∂¯ζ ′ ∧ det1,n−2 , ∂ ζ Φ∗ Φ Φ j=1 ∗ ′ χΦ w ¯ χw′ , ∂ζ ′ +(n − 1)∂¯ζ ′ . ∧ det1,n−2 Φ∗ Φ Φ
+(n − 1)
n−1
On the other hand we have ∗ ′ χwj w ¯ χw′ ′ , ∂ ∂¯ζ ′ ∧ det1,n−2 ζ Φ∗ Φ Φ ∗ w χwσ(n−1) χwj σ(1) ¯ χwσ(2) sgn(σ) ∧ ∧ · · · ∧ ∂¯ζ ′ = ∂¯ζ ′ ∂ζ ′ ∗ Φ Φ Φ Φ σ(1)=j
Integral Formulas for Strictly Pseudoconvex Domains
161
∗
χwn−1 wj ¯ χw1 ∧ · · · ∧ ∂¯ζ ′ ∂ζ ′ Φ∗ Φ Φ ′ ∗ χw wj = (n − 2)! ∗ ωζ ′ . Φ Φ
= (n − 2)!
Hence we have for ζ ∈ V
w(z, ζ) ¯ χ(ζ)w(z, ζ) zn (−1) det1,n−1 , ∂ζ ′ ζ) Φ∗ (z, ζ) Φ(z, ′ χw = −(n − 1)!ωζ ′ Φ ∗ ′ χΦ w ¯ χw′ ¯ ′ ′ +(n − 1)∂ζ . ∧ det1,n−2 , ∂ζ Φ∗ Φ Φ n
Now we set I(z) =
V
∗
χΦ ∧ det1,n−2 f (ζ)∂¯ζ ′ Φ
∗
w′ ¯ χw′ , ∂ζ ′ Φ∗ Φ
∧ ωζ ′ (ζ).
Since ∗ w(z, ζ), Φ∗ (z, ζ) and f (ζ) are holomorphic with respect to ζ, it follows from Stokes’ theorem that ∗ ′ w ¯ χw′ χΦ ¯ ∂ζ ′ f (ζ) ∧ ωζ ′ (ζ) ∧ det1,n−2 , ∂ζ ′ I(z) = Φ∗ Φ Φ V ∗ ′ w ¯ χw′ χΦ = ∧ ωζ ′ (ζ). det1,n−2 , ∂ζ ′ f (ζ) Φ∗ Φ Φ ∂V
ζ) and χ(ζ) = 1 for ζ ∈ ∂Ω, it follows from Stokes’ Since Φ(z, ζ) = Φ(z, theorem that ∗ ′ w ¯ χw′ ′ ∧ ωζ ′ (ζ) = 0, I(z) = ∂¯ζ ′ f (ζ)det1,n−2 , ∂ ζ Φ∗ Φ V which completes the proof of Lemma 3.22.
Lemma 3.23 There exists a constant C > 0 such that for all z ∈ ∂Ω\V the following estimates hold: (a)
ζ∈V
dVn−1 (ζ) C . ≤ |ζ − z|2n−1 |zn |
Several Complex Variables and Integral Formulas
162
(b)
dVn−1 (ζ)
ζ∈V ∩U2
(c)
ζ∈V ∩U2
ζ)| |Φ∗ (z, ζ)||ζ − z|2n−4 |Φ(z,
≤
C . |zn |
dVn−1 (ζ) C ≤ , 2 ∗ 2n−5 |z n| |Φ(z, ζ)| |Φ (z, ζ)||ζ − z|
where dVn−1 denotes the Lebesgue measure on Cn−1 .
Proof. In what follows we denote by C any constant which depends only on Ω and V . (a) we have dVn−1 (ζ) dVn−1 (ζ) ≤ . 2n−1 2 ′ ′ 2 2 ′ ′ 2n−5 ζ∈V |ζ − z| V (|zn | + |ζ − z | ) |ζ − z | We set ζj − zj = tj + itj+n−1 for j = 1, · · · , n − 1. Then dt1 · · · dt2n−2 dVn−1 (ζ) ≤ 2n−1 2 2 2 2n−5 |t|≤C (|zn | + |t| ) |t| ζ∈V |ζ − z| r2n−3 dr C . ≤ ≤ 2 2 )2 r2n−5 (|z | + r |z n n| r≤C This proves (a). (b) We set ζ ′ = (ζ1 , · · · , ζn−1 ) and z ′ = (z1 , · · · , zn−1 ). Let z 0 ∈ ∂V . We may assume that (∂ρ/∂z1 )(z 0 ) = 0. Let U be a neighborhood of z 0 such that (∂ρ/∂z1 )(z) = 0 for z ∈ U . For z, ζ ∈ U , define t2j−1 (ζ) = Re (ζj − zj ),
t2j (ζ) = Im (ζj − zj )
t1 (ζ) = ρ(ζ) − ρ(z),
j = 2, · · · , n − 1,
t2 (ζ) = Im Φ(z, ζ).
Then 1 ∂ρ ∂t2 (z) = − (z), ∂x2j 2 ∂x2j−1
∂t2 1 ∂ρ (z) = (z). ∂x2j−1 2 ∂x2j
Consequently, ∂ρ 2 ∂(t1 , · · · , t2n−2 ) = 0 = 2 ∂(x1 , · · · , x2n−2 ) ∂ζ1
Hence t1 , · · · , t2n−2 form a local coordinate system in U .
Integral Formulas for Strictly Pseudoconvex Domains
163
Since ρ(z) = 0, it follows from (3.19) that |Φ∗ (z, ζ)| ≥ |Φ(ζ, z)| ≥ C|F (ζ, z)| ≥ C(|t1 | + |ζ − z|2 ), ζ)| ≥ C(|t1 | + |ζ − z|2 ). |Φ(z,
Hence we have
dVn−1 (ζ) ζ)| |Φ∗ (z, ζ)||ζ − z|2n−4 |Φ(z, dt1 · · · dt2n−2 . (|zn |2 + |t1 | + |t|2 )2 |t|2n−4
ζ∈X∩Ω∩U2
≤
|t|≤C
We set t′ = (t2 , · · · , t2n−2 ). Then dt1 · · · dt2n−2 dt2 · · · dt2n−2 ≤ 2 2 2 2n−4 2 ′ 2 ′ 2n−4 |t|≤C (|zn | + |t1 | + |t| ) |t| |t′ |≤C (|zn | + |t | )|t | ∞ C dy C . ≤ ≤ 2 |zn | 0 1 + y |zn | This proves (b). (c) We have |Φ∗ (z, ζ)| = |Φ(ζ, z)| ≥ C(|t1 | + |t2 | + |ζ − z|2 ), ζ)| ≥ C(|t1 | + |t2 | + |ζ − z|2 ). |Φ(z,
We set t′′ = (t3 , · · · , t2n−2 ). Then we have dVn−1 (ζ) 2 ζ∈V ∩U2 |Φ(z, ζ)| |Φ∗ (z, ζ)||ζ − z|2n−5 dt1 · · · dt2n−2 ≤ 2 2 3 2n−5 |t|≤C (|zn | + |t1 | + |t2 | + |t| ) |t| dt3 · · · dt2n−2 ≤ 2 ′′ 2 3 ′′ 2n−5 |t′′ |≤C (|zn | + |t | ) |t | C . ≤ |zn | This proves (c).
Theorem 3.14 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let X = {z ∈ Cn | zn = 0}, V = Ω ∩ X. Assume that X
Several Complex Variables and Integral Formulas
164
intersects ∂Ω transversally. If f is a bounded holomorphic function in V , then there exists a constant C > 0 such that |Ef (z)| ≤ C sup |f (ζ)| ζ∈V
for z ∈ ∂Ω\V . We set U3 = {z ∈ U | |ρ(z)| < ε0 }. Since χ = 1 in U3 , we have
Proof.
det1,n−1
∗
w ¯ χw , ∂ζ ′ Φ∗ Φ
= det1,n−1
w ∂¯ζ ′ w ∂¯ζ ′ Φ − w , 2 Φ∗ Φ Φ
∗
. ∂¯ ′ Φ
ζ It follows from Lemma 3.4 that any determinant which contains w Φ 2 in two columns equals 0. Hence we have ∗ w ¯ χw ′ , ∂ζ det1,n−1 Φ∗ Φ ∗ ∗ ∂¯ζ ′ w w w ∂¯ζ ′ w ∂¯ζ ′ Φ + c2 det1,1,n−2 . = c1 det1,n−1 , , ,w 2 Φ∗ Φ∗ Φ Φ Φ
Since
∗
w(z, ζ) + w(z, ζ) = −w(ζ, z) + w(z, ζ) = O(|ζ − z|),
we have ∂¯ζ ′ w w ∂¯ζ ′ Φ det1,1,n−2 , ,w 2 Φ∗ Φ Φ ∂¯ζ ′ w ∂¯ζ ′ Φ O(|ζ − z|) . = det1,1,n−2 , ,w 2 Φ∗ Φ Φ
∗
It follows from (3.19) that
ζ)| ≥ α|ζ − z|2 , |Φ(z,
|Φ∗ (z, ζ)| ≥ α|ζ − z|2 .
Hence we obtain ∗ C C w ∂¯ζ ′ w det1,n−1 , ≤ ∗ n−1 ≤ ∗ Φ∗ Φ |Φ | |Φ| |Φ | |Φ| |ζ − z|2n−4
Integral Formulas for Strictly Pseudoconvex Domains
and
det1,1,n−2
≤
e ∂¯ζ ′ w ∂¯ζ ′ Φ O(|ζ − z|) ,w , ∗ e2 e Φ Φ Φ
C C|ζ − z| ≤ . ∗ n ∗ 2 e e |Φ | |Φ| |Φ | |Φ| |ζ − z|2n−5
Using Lemma 3.23, we have the desired inequality.
165
!
Theorem 3.15 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let X = {z ∈ Cn | zn = 0}, V = Ω ∩ X. Assume that X intersects ∂Ω transversally. If f is a bounded holomorphic function in V , then Ef is a bounded holomorphic function in Ω satisfying Ef = f on V . Moreover, there exists a constant C > 0 such that sup |F (z)| ≤ C sup |f (z)|. z∈Ω
z∈V
Proof. By Theorem 3.13, Ef is holomorphic in Ω ∪ U∂Ω\X . Let Xa = {z ∈ Cn | zn = a}. If a 6= 0, then Ef is holomorphic in the closure of Ω∩Xa . Hence by the maximum principle, |Ef | has the maximum in ∂(Ω ∩ Xa ). It follows from Theorem 3.14 that |Ef (z)| ≤ Ckf kV for z ∈ ∂(Ω ∩ Xa ), which means that |Ef (z)| ≤ Ckf kV for z ∈ Ω\X. Since Ef is holomorphic in Ω, |Ef (z)| ≤ Ckf kV for z ∈ Ω\(∂Ω ∩ X). Hence Ef is bounded in Ω. Next we prove bounded and continuous extensions of holomorphic functions from submanifolds in general position of strictly pseudoconvex domains in Cn with smooth boundary by the method of Henkin [HEN3]. For simplicity, we assume that the codimension of submanifolds is one. The general case can be proved in the same way. Definition 3.20 Let D be an open set in a complex manifold. We denote by H ∞ (D) the Banach space of all bounded holomorphic functions in D. We also denote by A(D) the Banach space of all continuous functions on D that are holomorphic in D. Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and e of Ω. We set X = let H be a holomorphic function in a neighborhood Ω e {z ∈ Ω | H(z) = 0} and V = X ∩ Ω. Assume that X intersects ∂Ω transversally. By Lemma 3.16, there are holomorphic functions h1 , · · · , hn in a neighborhood of Ω × Ω such that for z, ζ ∈ Ω H(z) − H(ζ) =
n X i=1
(zi − ζi )hi (z, ζ).
Several Complex Variables and Integral Formulas
166
By Theorem 3.10 there exist C ∞ functions w1 (z, ζ), · · · , wn (z, ζ) in a neighborhood of Ω × ∂Ω holomorphic with respect to z such that Φ(z, ζ) =
n X
wj (z, ζ)(zj − ζj ),
j=1
where Φ(z, ζ) is the function defined in Theorem 3.8. Define n−2
}| { z α(z, ζ) = −(−1)n(n+1)/2 det(wj , hj , ∂¯ζ wj , · · · , ∂¯ζ wj ),
Define
n−1 −2 n z }| { X ∂H 2 ∂H det β(ζ) = (ζ), dζj , · · · , dζj . ∂ζj (ζ) ∂ζj j=1
K(z, ζ) = α(z, ζ) ∧ β(ζ). Then Stout [STO] and Hatziafratis [HAT1] proved the following theorem. We omit the proof. Theorem 3.16
Let f ∈ A(V ). Then
f (z) =
Z
f (ζ)
∂V
K(z, ζ) Φ(z, ζ)n−1
for all z ∈ V . Remark 3.4 Stout [STO] obtained the integral formula on submanifolds of one codimension, and then Hatziafratis [HAT1] extended the integral formula obtained by Stout to the formula on submanifolds of any codimension. In what follows, we prove bounded and continuous extensions by following Henkin [HEN3]. Let Ω = {z | ρ(z) < 0} be a strictly convex domain in Cn with C ∞ boundary and let X = {zn = 0}, V = X ∩ Ω. Assume that X intersects ∂Ω transversally. We may assume that 0 ∈ Ω. Let f be a bounded holomorphic function in V . It follows from Fatou’s theorem (see Stein [STE]) that there is a bounded measurable function f ∗ on ∂V such that f ∗ (ζ) = lim f (θζ) θ↑1
for almost all ζ ∈ ∂V . Then we have the following lemma.
Integral Formulas for Strictly Pseudoconvex Domains
Lemma 3.24
167
For z ∈ V , we have K(z, ζ) f (z) = . f ∗ (ζ) Φ(z, ζ)n−1 ζ∈∂V
Proof. Let 0 < θ < 1. We set F (z) = f (θz) for z ∈ V . Then F is holomorphic in a neighborhood of V . We fix z0 ∈ V . It follows from Theorem 3.16 that K(z0 , ζ) . F (ζ) F (z0 ) = n−1 Φ(z 0 , ζ) ζ∈∂V Consequently, z0 f( ) = θ
f (θζ)
ζ∈∂V
K(z0 , ζ) . Φ(z0 , ζ)n−1
By Lebesgue’s dominated convergence theorem, we may pass the limit under the integral sign as θ → 1 in the above equality. Definition 3.21 Let f be a bounded holomorphic function in V and let f ∗ be the boundary value of f on ∂V . For z ∈ Ω, define K(z, ζ) E1 f (z) = f ∗ (ζ) . (3.30) Φ(z, ζ)n−1 ∂V Then E1 f is holomorphic in a neighborhood of Ω\∂V and E1 f |V = f . If f ∈ A(V ), then f |∂V = f ∗ . Let Ω = {z | ρ(z) < 0} be a strictly convex domain in Cn with C ∞ boundary and let X = {zn = 0}, V = X ∩ Ω. Assume that X intersects ∂Ω transversally. We fix a point z ∗ ∈ ∂V . Suppose ∂ρ ∗ (z ) = 0. ∂z1 ∂ρ Then there exists a constant σ1 > 0 such that ∂z (z) = 0 for all z ∈ 1 ∗ B(z , σ1 ). In this setting, we prove Lemma 3.25, Lemma 3.26, Lemma 3.27, Theorem 3.17 and Theorem 3.18.
Lemma 3.25 For z ∈ B(z ∗ , σ1 ), we consider a system of equations for ζ ∗ = (ζ1∗ , · · · , ζn∗ ) of the following form: n ∂ρ ∗ ∗ (ζ )(ζi − zi ) = 0, ∂ζi i=1
(3.31)
Several Complex Variables and Integral Formulas
168
ζi∗ = zi
(i = 2, · · · , n − 1),
ζn∗ = 0.
(3.32)
Then there exist positive constants σ2 , γ1 and γ2 depending only on Ω and V , such that for any z ∈ B(z ∗ , σ2 ) there exists a unique solution ζ ∗ = ζ ∗ (z) of the system of equations (3.31) and (3.32) which belongs to the set B(z ∗ , σ2 ) ∩ X. Moreover, the point ζ ∗ has the following properties: |z − ζ ∗ |2 ≤
1 {ρ(z) − ρ(ζ ∗ )}, γ1
|z − ζ ∗ |2 ≥ |zn |2 ≥ γ2 {ρ(z) − ρ(ζ ∗ )}, ζ∗ = z Proof.
z ∈ B(z ∗ , σ2 ) ∩ X.
for
(3.31) can be written ∂ρ ∗ (ζ ) ∂zn
∂ρ g(ζ) = (ζ) ∂zn
ζ1∗ = z1 +
−1 ∂ρ ∗ (ζ ) zn . ∂z1
Define −1 ∂ (ζ) . ∂z1
We choose σ2 > 0 so small that |dg(ζ)||zn | ≤ 1/2 for z, ζ ∈ B(z ∗ , σ2 ). Define {ζ (j) } by recurrence such that (1)
= z1 ,
(j)
= (ζ1 , z2 , · · · , zn−1 , 0),
ζ1 ζ
(j+1)
ζ1 (j)
(j−1)
(j)
= z1 + g(ζ (j) )zn .
(j−1)
(j−2)
Then |ζ1 − ζ1 | ≤ 21 |ζ1 − ζ1 |, and hence {ζ (j) } converges. Let (j) ∗ ∗ limj→∞ ζ = ζ . Then ζ satisfies (3.31) and (3.32). The strict convexity of ρ yields for some positive constants γ1 and C1 ρ(ζ ∗ ) − ρ(z) + γ1 |ζ ∗ − z|2 ≤ 2Re
n ∂ρ ∗ ∗ (ζ )(ζj − zj ) = 0 ∂ζj j=1
n ∂ρ ∗ ∗ ρ(ζ ) − ρ(z) + C1 |ζ − z| ≥ 2Re (ζ )(ζj − zj ) = 0. ∂ζ j j=1 ∗
∗
2
Integral Formulas for Strictly Pseudoconvex Domains
169
Since ζ1∗ = z1 + g(ζ ∗ )zn , there exists a constant C2 such that |ζ1∗ − z1 | ≤ C2 |zn |. Hence ζ ∗ satisfies the desired inequalities. If there are two solutions ζ ∗ and ζ˜∗ . Then we have |ζ1∗ − ζ˜1∗ | ≤ dg|ζ1∗ − ζ˜1∗ ||zn | ≤ 1/2|ζ1∗ − ζ˜1∗ |, which implies that ζ ∗ = ζ˜∗ . Lemma 3.26 have
For any z ∗ ∈ ∂V and any z ∈ (∂Ω\∂V ) ∩ B(z ∗ , σ2 ) we d(E1 f )(ζ ∗ + λ(z − ζ ∗ )) dλ
λ=1
≤ C sup |f (ζ)| ζ∈V
where ζ ∗ = ζ ∗ (z) and σ2 are from Lemma 3.25 and the constant C depends only on Ω and V . Proof.
Since n ∂ρ ∗ ∗ (ζ )(ζi − zi ) = 0 ∂ζi j=1
and ∂Φ ∂ρ ∗ (z, ζ ∗ ) = −2 (ζ ) + O(|ζ ∗ − z|), ∂zi ∂ζi we have n n ∂ρ ∂Φ ∂Φ (z, ζ)(ζi∗ − zi ) = (z, ζ) + 2 (ζ ∗ ) (ζi∗ − zi ) ∂zi ∂zi ∂ζi i=1
i=1
n ∂Φ ∂Φ ∗ ∗ ∗ ≤ (z, ζ) − (z, ζ ) + O(|ζ − z|) (ζi − zi ) ∂z ∂z i i i=1 ≤ Cε(|ζ − z| + ε),
where ε = |zn |. Then we have d(E1 f )(ζ ∗ + λ(z − ζ ∗ )) dλ ≤C
+C
∂V ∩B(z ∗ ,σ2 )
∂V ∩B(z ∗ ,σ2 )
λ=1
d(E1 f )(z + λ(z − ζ ∗ )) = dλ
|f ∗ (ζ)||z − ζ ∗ | dσ2n−3 (ζ) |Φ(z, ζ)|n
|f ∗ (ζ)|ε(|ζ − z| + ε) dσ2n−3 (ζ). |Φ(z, ζ)|n
λ=0
170
Several Complex Variables and Integral Formulas
We choose a local coordinate system (t1 , t2 , · · · , t2n ) in B(z ∗ , σ2 ) such that t1 + it2 = ρ(ζ) − ρ(z) + iIm Φ(z, ζ). Then we obtain d(E1 f )(ζ ∗ + λ(z − ζ ∗ )) ≤ C sup |f (ζ)|. dλ ζ∈V λ=1
Theorem 3.17 If f ∈ H ∞ (V ), then E1 f ∈ H ∞ (Ω). Moreover, E1 : H ∞ (V ) → H ∞ (Ω) defines a bounded linear operator. Proof. Let σ > 0. Let f be a bounded holomorphic function in V . Since E1 f is holomorphic in Ω\∂V , it is sufficient to prove that sup z∈∂Ω\∂V
|E1 f (z)| ≤ C sup |f (ζ)|. ζ∈V
It is easily proved that supz∈∂Ω\(∂V )σ |E1 f (z)| ≤ C supζ∈V |f (ζ)|, where (∂V )σ is the σ-neighborhood of ∂V . Therefore, it is sufficient to show that sup z∈{(∂V )σ \∂V }∩∂Ω
|E1 f (z)| ≤ C sup |f (ζ)|. ζ∈V
Let z ∈ {(∂V )σ \∂V } ∩ ∂Ω. We set ∆ = {λ ∈ C | z(λ) = ζ ∗ + λ(z − ζ ∗ ) ∈ Ω}. Then ∆ is a convex domain containing λ = 0 since Ω is convex. Since ρ(z) = 0, we have by Lemma 3.25 ε2 ε2 ≤ −ρ(ζ ∗ ) ≤ . γ1 γ2 If λ ∈ ∂∆, then z(λ) ∈ ∂Ω, and hence ρ(z(λ)) = 0. Then
ε 1 & ρ(z) − ρ(ζ ∗ ) ≤ √ |z(λ) − ζ ∗ | ≤ √ γ1 γ1 γ2
for λ ∈ ∂∆. Consequently,
|z(λ) − z ∗ | ≤ |z(λ) − ζ ∗ | + |ζ ∗ − z ∗ | ≤√
σ σ2 + . γ1 γ2 4
√ We impose the further assumption that the constant σ < σ2 γ1 γ2 /4. Then |z(λ)− z ∗ | < σ2 /2. Then ζ ∗ = ζ ∗ (z) satisfies (3.31) and (3.32) for z(λ) with
Integral Formulas for Strictly Pseudoconvex Domains
171
λ ∈ ∂∆, and hence ζ(z(λ)) = ζ ∗ (z) for any λ ∈ ∂∆. Moreover, it follows from Lemma 3.25 that |λ| & ε |λ|ε ≥ √ ρ(z) − ρ(ζ ∗ ) ≥ |λ||z − ζ ∗ | = |z(λ) − ζ ∗ | ≥ ε √ γ1 γ2 γ1 for λ ∈ ∂∆. Consequently, |λ| ≥ Since
√ γ1 γ2
for any λ ∈ ∂∆.
d(E1 f )(ζ ∗ + λ(z − ζ ∗ )) d(E1 f )(ζ ∗ + t(z(λ) − ζ ∗ )) = λ, dt dλ t=1
we obtain for some constant C1 > 0 and C2 > 0, d(E1 f )(ζ ∗ + λ(z − ζ ∗ )) C1 ≤ sup |f (ζ)| ≤ C2 sup |f (ζ)| |λ| ζ∈V dλ ζ∈V
for every λ ∈ ∂∆. Since the function d(E1 f )(ζ ∗ + λ(z − ζ ∗ ))/dλ is holomorphic for all λ ∈ ∆, we have for some constant C3 > 0 d(E1 f )(ζ ∗ + λ(z − ζ ∗ )) ≤ C3 sup |f (ζ)|. sup dλ λ∈∆ ζ∈V
Consequently, there exists a constant C4 > 0 such that 1 d E1 f (ζ ∗ + λ(z − ζ ∗ ))dλ ≤ C4 sup |f (ζ)|. |E1 f (z) − E1 f (ζ ∗ )| = ζ∈V 0 dλ
Since ζ ∗ ∈ V , we have E1 f (ζ ∗ ) = f (ζ ∗ ). Hence there exists a constant C5 > 0 such that sup z∈{(∂V )σ \∂V }∩∂Ω
|E1 f (z)| ≤ C5 sup |f (ζ)|, ζ∈V
which completes the proof of Theorem 3.17.
Lemma 3.27 Let V ′ be a domain with smooth boundary in X such that V ⊂ V ′ . We denote by C any positive constant which depends only on Ω, V and V ′ . Then for z ∈ Ω and ε = |zn |, (a)
V
′ \V
1 dVn−1 (ζ) ≤ C| log ε|. |Φ(z, ζ)|n
Several Complex Variables and Integral Formulas
172
(b) Z
V
′ \V
|z − ζ| 1 dVn−1 (ζ) ≤ C . |Φ(z, ζ)|n+1 ε
(c) Z
V ′ \V
|z − ζ|2 dVn−1 (ζ) ≤ C| log ε|. |Φ(z, ζ)|n+1
Proof. We may assume that |z − ζ| < ε, where ε is the constant in Theorem 3.8. By contracting V ′ if necessary, it follows from (3.19) that 2ReΦ(z, ζ) ≥ ρ(ζ) − ρ(z) + β|ζ − z|2 for (z, ζ) ∈ {Ω × (V ′ \V )} ∩ {(z, ζ) | |ζ − z| < ε}. We can choose a local coordinate system t = (t1 , · · · , t2n−2 ) such that ρ(ζ) = t1 , Im Φ(z, ζ) = t2 . We set t′ = (t3 , · · · , t2n−2 ). Then (a) Z Z dt1 · · · dt2n−2 1 dVn−1 (ζ) ≤ C n ′ |2 + ε2 + |t | + |t |)n |Φ(z, ζ)| (|t ′ 1 2 |t|≤R V \V Z dt3 · · · dt2n−2 ≤C ′ 2 2 n−2 |t′ |≤R (|t | + ε ) Z R r2n−5 ≤C dr 2 2 n−2 0 (r + ε ) ≤ C| log ε|. (b) Z
V ′ \V
Z
|z − ζ| dVn−1 (ζ) |Φ(z, ζ)|n+1
1 dVn−1 (ζ) + |Im Φ(z, ζ)| + |ρ(ζ)|)n+(1/2) V ′ \V (|z − Z dt3 · · · dt2n−2 1 ≤C ≤C . ′ |2 + ε2 )n−(3/2) ε (|t ′ |t |≤R
≤C
ζ|2
(c) Z
V
′ \V
|z − ζ|2 dVn−1 (ζ) ≤ C |Φ(z, ζ)|n+1
Z
|t′ |≤R
dt3 · · · dt2n−2 ≤ C| log ε|. (|t′ |2 + ε2 )n−2
Integral Formulas for Strictly Pseudoconvex Domains
Theorem 3.18 If f ∈ A(V ), then E1 f ∈ A(Ω). A(V ) → A(Ω) defines a bounded linear operator.
173
Moreover, E1 f :
Proof. Let z ∗ ∈ ∂V and let z ∈ (Ω\V )∩B(z ∗ , σ2 ). Let ζ ∗ be the solution for the system of equations (3.31) and (3.32). We set z(θ) = ζ ∗ + θ(z − ζ ∗ ) for 0 ≤ θ ≤ 1. Then z(θ)n = θzn . Further, ζ ∗ = ζ ∗ (z) also satisfies the system of equations (3.31) and (3.32) for z(θ) instead of z. By the uniqueness of the solution, we have ζ ∗ (z) = ζ ∗ (z(θ)). Let V ′ be a domain with smooth boundary in X such that V ⊂ V ′ . Then f (ζ)K(z, ζ) E1 f (z) = Φ(z, ζ)n−1 ∂V f (ζ)K(z, ζ) = n−1 ′ ∂V Φ(z, ζ) K(z, ζ) ¯ f (ζ)∂ζ − . Φ(z, ζ)n−1 V ′ \V Define F1 (z) =
f (ζ)∂¯ζ
(V ′ \V )∩B(z ∗ ,σ2 )
K(z, ζ) Φ(z, ζ)n−1
.
By Theorem 3.8, we may assume that Φ(z, ζ) = F (z, ζ)M (z, ζ) on B(z ∗ , σ2 ) × B(z ∗ , σ2 ), where F (z, ζ) is the Levi polynomial. Then F1 is expressed by A(z, ζ) f (ζ) F1 (z) = Φ(z, ζ)n−1 ′ ∗ (V \V )∩B(z ,σ) n j=1 (ζj − zj )Bj (z, ζ) f (ζ) + , Φ(z, ζ)n (V ′ \V )∩B(z ∗ ,σ) where A(z, ζ) and Bj (z, ζ), 1 ≤ j ≤ n, are C ∞ (2n − 2) forms on Ω × V . Then it follows from Lemma 3.27 that there exist constants C1 , C2 and C3 such that dF1 (ζ ∗ + λ(z − ζ ∗ )) ε ≤ C1 |f | | dVn−1 (ζ) λ=1 V n dλ (V ′ \V )∩B(z ∗ ,σ) |Φ(z, ζ)| +C2 |f |V
(V ′ \V )∩B(z ∗ ,σ)
ε|ζ − z|(|ζ − z| + ε) dVn−1 (ζ) ≤ C3 ε| log ε|f |V , |Φ(z, ζ)|n+1
for any point z ∗ ∈ ∂V ,z ∈ (Ω\∂V ) ∩ B(z ∗ , σ2 ), where ε = |zn |.
Several Complex Variables and Integral Formulas
174
Then ∗
|F1 (z) − F1 (ζ )| = = = ≤
1 d ∗ ∗ F1 (ζ + θ(z − ζ ))dθ 0 dθ 1 dF1 (ζ ∗ + λθ(z − ζ ∗ ) |λ=1 dθ dλ 0 1 ∗ 1 dF1 (ζ + λ(z(θ) − ζ ∗ )) dθ dλ 0 θ λ=1 1 ε| log εθ|dθ sup |f (ζ)|. C4 0
ζ∈V
Hence we obtain |E1 f (z) − E1 f (ζ ∗ )| ≤ C5 σ2 | log σ2 | sup |f (ζ)| ζ∈V
We may assume that f ∈ C 1 (V ). Since ζ ∗ ∈ V , we obtain |E1 f (z) − E1 f (z ∗ )| = |E1 f (z) − E1 f (ζ ∗ )| + |f (ζ ∗ ) − f (z ∗ )|
≤ |E1 f (z) − E1 f (ζ ∗ )| + C6 {|ζ ∗ − z| + |z − z ∗ |}
≤ C7 σ2 | log σ2 | sup |f (ζ)|. ζ∈V
Consequently, limz→z∗ E1 f (z) = f (z ∗ ).
Lemma 3.28 Let Ω be a strictly pseudoconvex domain with C ∞ boundary in Cn and let X be a closed submanifold of codimension one in a neighbor of Ω. Let Ω′ be a pseudoconvex domaim such that Ω ⊂ Ω′ ⊂ Ω′ ⊂ Ω. hood Ω ′ ′ We set V = Ω ∩ X and V = Ω ∩ X. Assume that X intersects ∂Ω transversally. Let ζ ∈ ∂V . It follows from Theorem 5.20 (a) (Cartan theorem A) that there exist σ > 0 and holomorphic functions F1 , · · · , Fq ∈ Γ(Ω′ , FV ′ ) such that FV ′ is generated by F1 , · · · , Fq in B(ζ, σ). Then there exist constants σ > σ1 > δ > 0 with the following properties: (a) For some integer q1 with 1 ≤ q1 ≤ q and some integers m1 , · · · , mn−2 from the set {1, · · · , n} the mapping ϕ(z) = (zm1 − ζm1 , · · · , zmn−2 − ζmn−2 , F (z, ζ), Fq1 (z)) is a biholomorphic mapping of the ball B(ζ, σ1 ) ⊂ Ω′ onto a neighborhood Wζ of 0, where F (z, ζ) is the Levi polynomial defined in Definition 3.12.
Integral Formulas for Strictly Pseudoconvex Domains
175
(b) There exists a strictly convex domain Uζ ⊂⊂ Wζ such that Ω ∩ B(ζ, δ) ⊂ ϕ−1 (Uζ ) ⊂ Ω, where Uζ = {w ∈ Wζ | ρζ (w) < 0}, and ρζ is a real-valued C 2 function in the domain Wζ that is strictly convex in a neighborhood of U ζ . Proof.
There exists q1 (1 ≤ q1 ≤ q) such that the equation m ∂Fq
1
j=1
∂zj
(ζ)(zj − ζj ) = 0
defines a (n − 1) dimensional analytic plane tangent to V ′ at the point z = ζ. Since X and ∂Ω intersect transversally, the equations m ∂Fq
1
j=1
∂zj
(ζ)(zj − ζj ) = 0,
n n ∂F ∂ρ (ζ, ζ)(zj − ζj ) = 2 (ζ)(zj − ζj ) = 0 ∂zj ∂zj j=1 j=1
define a (n − 2) dimensional analytic plane if ζ ∈ ∂V . Therefore we can choose numbers m1 , · · · , mn−2 so that ϕ(z) = {zm1 − ζm1 , · · · , zmn−2 − ζmn−2 , F (z, ζ), Fq1 (z)) has a non-zero Jacobian at the point z = ζ. By the implicit function theorem there exists σ > 0 such that the mapping ϕ of the ball B(ζ, σ) onto some domain Wζ containing 0 has the inverse mapping ϕ−1 (see Corollary 5.3). Define ρζ (w) = ρ(ϕ−1 (w)). By Taylor’s formula we have n n 2 ∂ρ ∂ ρ ρ(z) = Re 2 (ζ)(zj − ζj ) + (ζ)(zj − ζj )(zk − ζk ) ∂z ∂z ∂z j j k j=1 j,k=1
+
n
j,k=1
2
∂ ρ (ζ)(zj − ζj )(¯ zk − ζ¯k ) + o(|z − ζ|2 ). ∂zj ∂ z¯k
Since F (z, ζ) = wn−1 and zi − ζi = zi (w) − zi (0) =
n ∂zi (0)wν + o(|w|), ∂wν ν=1
Several Complex Variables and Integral Formulas
176
we obtain ρζ (w) = Rewn−1 +
n
i,j=1
n ∂zi ∂2ρ (ζ) (0)wν ∂zi ∂ z¯j ∂wν ν=1
n ∂ z¯j (0)w ¯ν × ∂ w ¯ν ν=1
+ o(|w|2 ).
ρζ is strictly convex in a neighborhood U1 (⊂ U0 ) of w = 0. Define tζ (w) = Re
n ∂ ρζ (0)wi . ∂w i i=1
The equation tζ (w) = 0 defines the real tangent plane to the boundary of the convex domain U2 := {w ∈ U1 | ρζ (w) < 0} at the point w = 0. Since U2 is strictly convex near 0, there exists ε > 0 such that if we define U3 = {w ∈ U1 | ρζ (w) < 0, tζ (w) > −ε}, then U3 ⊂⊂ U1 . Define 0 (t ≥ − 2ε ) . χ(t) = (t + 2ε )4 (t ≤ − 2ε ) Then χ is of class C 2 in R. We choose a constant A > 0 in such a way that ρζ (w)| < Aχ(−ε). sup sup |
ζ∂V w∈U1
Define ρζ (w) = ρζ (w) + Aχ(tζ (w)).
Since 2n
j,k=1
2 2n ∂ 1 ∂ ρζ [χ(tζ (w))]uj uk = χ′′ (tζ (w)) (0)uj ≥ 0, ∂uj ∂uk 4 ∂u j j=1 2
ρζ (w) is strictly convex in U1 . Then Uζ = {w ∈ U1 | ρζ (w) < 0} is a strictly convex domain in U1 with Uζ ⊂⊂ U1 . Define Gζ = {z ∈ B(ζ, σ) | ϕ(z) ∈ Uζ }. If we choose δ > 0 sufficiently small, then we obtain Ω ∩ B(ζ, δ) ⊂ Gζ ⊂ Ω. Lemma 3.29 Let LΩ and RΩ be the integral operators defined in Definition 3.14 and Definition 3.16, respectively. (a) If f is a bounded function in Ω, then RΩ f is continuous on Ω.
Integral Formulas for Strictly Pseudoconvex Domains
177
(b) If f is a bounded holomorphic function in Ω and if ϕ is a C 1 function in Cn , then LΩ (f ϕ) is bounded in Ω. Proof.
(a) We write RΩ f in the following form RΩ f (z) = f (ζ)H(z, ζ)dV (ζ), Ω
where dV (ζ) is the Lebesgue measure on Ω. Then 1 ζ¯ − z¯ χ(ζ)w(z, ζ) +λ |H(z, ζ)| ≤ C ω 1 − λ dλ ζ) 0 |z − ζ|2 Φ(z, 1 1 1 . + + ≤C 2 |ζ − z|2n−3 − z|2n−2 |ζ − z|2n−1 |Φ| |Φ||ζ For ε > 0, there exists δ > 0 such that for any w ∈ Ω |H(w, ζ)|dV (ζ) < ε. B(w,3δ)∩Ω
Let Kδ = {(w, ζ) ∈ Ω × ∂Ω | |w − ζ| ≥ δ}. Since H(w, ζ) is continuous on the compact set Kδ , we can choose δ > δ1 > 0 such that for any point w with |z − w| < δ1 , |H(z, ζ) − H(w, ζ)| < ε for all ζ satisfying |ζ − z| ≥ 2δ. For |z − w| < δ1 , we have |RΩ f (z) − RΩ f (w)| = f (ζ)(H(z, ζ) − H(w, ζ))dV (ζ) Ω
≤ (H(z, ζ) − H(w, ζ))dV (ζ) Ω\B(z,2δ)
Ω∩B(z,2δ)
|f (ζ)||H(z, ζ)|dV (ζ)
Ω∩B(w,3δ)
|f (ζ)||H(w, ζ)|dV (ζ)
+
+
Several Complex Variables and Integral Formulas
178
≤ε
Ω\B(z,2δ)
|f ζ)|dV (ζ) + εf + εf < Cε.
Hence RΩ f is continuous on Ω. ¯ ϕ) = f ∂ϕ, ¯ ∂(f ¯ ϕ) is bounded in Ω. It follows from Theorem (b) Since ∂(f 3.12 that ¯ )), ϕf = LΩ (ϕf ) + RΩ (∂(ϕf which means that LΩ (ϕf ) is bounded in Ω.
Lemma 3.30 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let Uj ⊂ Cn (j = 1, · · · , N ) be open sets such that Ω ⊂ N ∞ ∪N j=1 Uj . Then any f ∈ H (Ω) admits a decomposition f = j=1 fj , where every fj is bounded and holomorphic in some neighborhood of Ω\(∂Ω ∩ Uj ). In addition, if f is continuous on Ω, then every fj is continuous on Ω. N Proof. Choose C ∞ functions χj in Cn such that j=1 χj = 1 on Ω and supp(χj ) ⊂ Uj . Define fj = LΩ (χj f ). Since χj = 0 on Cn \Uj , fj is bounded and holomorphic in some neighborhood of Ω\(∂Ω ∩ Uj ). By Theorem 3.12 f = LΩ f =
N j=1
LΩ (χj f ) =
N
fj .
j=1
Suppose f is continuous on Ω. By Theorem 3.12 we have ¯ j ). fj = χj f − RΩ (f ∂χ ¯ j ) is continuous on Ω. Hence fj is continuous on By Lemma 3.29 RΩ (f ∂χ Ω. The following theorem was proved by Henkin [HEN3] which is a generalization of Lemma 3.30 to submanifolds of strictly pseudoconvex domains in Cn with smooth boundary. Theorem 3.19 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with of C ∞ boundary and let X be a closed submanifold in a neighborhood Ω Ω, V = Ω ∩ X. Assume that X intersects ∂Ω transversally. Let Uj ⊂ X (j = 1, · · · , N ) be open sets in X such that V ⊂ ∪N j=1 Uj . Then any f ∈ N ∞ H (V ) admits a decomposition f = j=1 fj , where every fj is bounded and holomorphic in some neighborhood of V \(∂V ∩ Uj ). In addition, if f is continuous on V , then every fj is also continuous on V .
Integral Formulas for Strictly Pseudoconvex Domains
179
Proof. Let Ω′ be a strictly psuedoconvex domain such that Ω ⊂⊂ Ω′ ⊂⊂ We set V ′ = Ω′ ∩ X. Let ε > 0 be given. Let χi , i = 1, · · · , N , be realΩ. N valued, nonnegative C ∞ functions such that i=1 χi = 1 in a neighborhood ′ of V and the diameter of each set supp(χi ) is less that ε/3. Define χi , χ1ν = {i | supp(χi )∩supp(χν )=φ} χ ν =
{i |
χi .
supp(χi )∩supp(χ1ν )=φ}
We consider domains for ν = 1, · · · , N , ′
Ων = {z ∈ Ω | ρ(z) −
ν = {z ∈ Ω′ | ρ(z) − Ω
ν−1 i=1
ν
λi χi (z) < 0},
i=1
λi χi (z) − λ χν (z) < 0}.
ν for ν = 0, 1, · · · , N . In We set Ω0 = Ω, Vν = V ′ ∩ Ων and Vν = V ′ ∩ Ω order to prove Theorem 3.19 we need the following lemma. Lemma 3.31 For sufficiently small λ1 , · · · , λN > 0 and for any ν = 1, · · · , N , there exist bounded operators L0ν : H ∞ (Vv−1 ) → H ∞ (Vν ) and L1ν : H ∞ (Vν−1 ) → H ∞ (Vν ) with the following properties:
(a) f (z) = (L0ν f )(z) + (L1ν f )(z) for any f ∈ H ∞ (Vν−1 ) and any z ∈ Vν−1 . (b) L0ν f ∈ A(Vν ) and L1ν f ∈ A(Vν ) if f ∈ A(Vν−1 ).
Proof of Lemma 3.31. Suppose that constants λ1 , · · · , λν−1 satisfying the conditions of the lemma have already been chosen. We set Uν = supp(χ1ν ). We may assume that Uν ∩ ∂Vν−1 = φ. We fix a point ζ ∗ ∈ Uν ∩ ∂Vν−1 . By Lemma 3.28, there exists a biholomorphic mapping ϕ of the ball B(ζ ∗ , (3/4)σ) onto a neighborhood Wζ ∗ of 0 such that Ων−1 ∩ B(ζ ∗ , (3/4)δ) ⊂ Gζ ∗ = {z ∈ B(ζ ∗ , (3/4)δ) | ρζ ∗ (ϕ(z)) < 0},
where ρζ ∗ (w) is strictly convex in a neighborhood of the set E ζ ∗ = {w ∈ Wζ ∗ | ρζ ∗ (w) ≤ 0}. We set Iζ ∗ = E ζ ∗ ∩ ϕ(Vν−1 ∩ Gζ ∗ ). For any function f ∈ H ∞ (Vν−1 ) and any z ∈ Gζ ∗ ∩ Vν−1 , it follows from (3.30) that K(w, ζ) f (z) = f (ϕ−1 (w)) = . f (ϕ−1 (ζ)) Φ(w, ζ)n−1 ζ∈∂Iζ ∗
Several Complex Variables and Integral Formulas
180
We set χ0ν = 1 − χ1ν and Rνα f (z)
=
ζ∈∂Iζ ∗
−1 f (ϕ−1 (ζ))χα (ζ)) ν (ϕ
K(ϕ(z), ζ) Φ(ϕ(z), ζ)n−1
for α = 1, 2 and f ∈ H ∞ (Vν−1 ). Then we have f (z) = (Rνα f )(z) + (Rν1 f )(z)
(z ∈ Gζ ∗ ∩ Vν−1 ).
We choose λν < λ0 sufficiently small. We set V
′′
V0′′
′
= {z ∈ V | ρ(z) − ′′
ν−1 i=1
λi χi (z) < λ0 } = V0′′ ∪ V1′′ ,
where = V ∩ B(ζ , (3/4)δ) and V1′′ = V ′′ \B(ζ ∗ , (1/2)δ) . Then we have a representation K(ϕ(z), ζ) −1 (ζ)) f (ϕ−1 (ζ))χα Rνα f (z) = ν (ϕ Φ(ϕ(z), ζ)n−1 ζ∈∂Iζ ∗ = χα ν (z)f (z) +
∗
ζ∈∂Iζ ∗
−1 f (ϕ−1 (ζ)){χα (ζ)) − χα ν (ϕ ν (z)}
K(ϕ(z), ζ) . Φ(ϕ(z), ζ)n−1
We set Aα ν (z) =
ζ∈∂Iζ ∗
−1 f (ϕ−1 (ζ)){χα (ζ)) − χα ν (ϕ ν (z)}
K(ϕ(z), ζ) . Φ(ϕ(z), ζ)n−1
Then we can prove that A0ν is a bounded operator from H ∞ (Vν−1 ) to A(Vν ∩ B(ζ ∗ , (3/4)δ), and A1ν is a bounded operator from H ∞ (Vν−1 ) to A(Vν ∩ B(ζ ∗ , (3/4)δ) using the same method as the proof of Lemma 3.29 (a). Therefore, we can prove that Rν0 is a bounded operator from H ∞ (Vν−1 ) to H ∞ (Vν ∩ B(ζ ∗ , (3/4)δ). On the other hand Rν1 is a bounded operator from H ∞ (Vν−1 ) to H ∞ (Vν ∩ B(ζ ∗ , (3/4)δ). If we choose 0 < χν < χ0 sufficiently small, then χ1ν = 0 in a neighborhhood of V0′′ ∩ V1′′ . Hence Rν1 is a bounded operator from H ∞ (Vν−1 ) to A(V0′′ ∩ V1′′ ). If f ∈ A(Vν−1 ), then Rν0 f ∈ A(Vν ∩ B(ζ ∗ , (3/4)δ)) and Rν1 f ∈ A(Vν ∩ B(ζ ∗ , (3/4)δ)).
Integral Formulas for Strictly Pseudoconvex Domains
181
It follows from Theorem 5.26 that for f ∈ O(V0′′ ∩ V1′′ ) there exist mappings Tνα : O(V0′′ ∩ V1′′ ) → O(Vα′′ ) such that f = Tν0 f + Tν1 f. For z ∈ Vν−1 ∩ V0′′ ∩ V1′′ , we have f (z) = Rν0 f (z) + Rν1 f (z) = Rν0 f (z) + Tν0 (Rν1 f )(z) + Tν1 (Rν1 f )(z). We set (L0ν f )(z)
(L1ν f )(z)
=
=
(
(Rν0 f )(z) + (Tνo ◦ Rν1 f )(z) (z ∈ Vν ∩ B(ζ ∗ , (3/4)δ)) , f (z) − (Tν1 ◦ Rν1 f )(z) (z ∈ Vν \B(ζ ∗ , (1/2)δ)) (Rν1 f )(z) + (Tνo ◦ Rν1 f )(z) (z ∈ Veν ∩ B(ζ ∗ , (3/4)δ)) , (Tν1 ◦ Rν1 f )(z) (z ∈ Veν \B(ζ ∗ , (1/2)δ))
Then L0ν and L1ν satisfy conditions (a) and (b), which completes the proof of Lemma 3.31. End of the proof of Theorem 3.19. We set Li = L1i ◦ L0i−1 ◦ · · · ◦ L01 for i = 1, · · · , N − 1, and LN = L0N −1 ◦ L0N −2 ◦ · · · L01 . If f ∈ H ∞ (V ), then Li f ∈ H ∞ (Vei ) for i = 1, 2, · · · , N −1, and LN f ∈ H ∞ (VN −1 ). If f ∈ A(V ), then Li f ∈ A(Vei ) for i = 1, 2, · · · , N − 1, and LN f ∈ A(VN −1 ). Moreover, if f ∈ H ∞ (V ) and z ∈ V , then f (z) =
N X
Li f (z).
i=1
The diameter of the set V \VeN −1 is less than ε, and the diameter of the set V \VN −1 is less than ε/3. Theorem 3.19 is proved.
Now we are going to prove bounded and continuous extensions of holomorphic functions from submanifolds in general position of a strictly pseudoconvex domain in Cn with C ∞ boundary.
Corollary 3.6 Let Ω be a strictly pseudoconvex domain in Cn with C ∞ boundary and let X be a closed submanifold in a neighborhood of Ω. Let V = Ω ∩ X. Assume that X intersects ∂Ω transversally. Then for any f ∈ H ∞ (V ) there exists g ∈ H ∞ (Ω) such that g = f on V . Moreover, if f ∈ A(V ), then there exists g ∈ A(Ω) such that g = f on V .
182
Several Complex Variables and Integral Formulas
Proof. Let f be a bounded holomorphic function in V . Since V is compact, there is a biholomorphic mapping hξ : B(ξ, δ) → Cn such that hξ (X ∩ B(ξ, δ)) is the intersection of hξ (B(ξ, δ)) with a complex hyperplane in Cn . By Theorem 3.19, it is sufficient to prove the theorem for the case when f has the following property: There is a point ξ ∈ ∂V and a strictly pseudoconvex open set Ω0 ⊂ Cn such that Ω\ (∂Ω ∩ B (ξ, δ/3)) ⊂ Ω0 and f is bounded holomorphic in X ∩ Ω0 . We can choose a strictly pseudoconvex open set Ωξ ⊂ Cn such that B (ξ, δ/3) ∩ Ωξ ⊂ B (ξ, δ/2) ∩ Ω0 . Then we have Ω ⊂ Ω0 ∪ B (ξ, δ/3) . Therefore, we can choose a strictly pseudoconvex open set Ω1 such that Ω ⊂⊂ Ω1 ⊂⊂ Ω0 ∪ B (ξ, δ/3) . We set Uξ := B (ξ, δ/3) ∩ Ω1 ,
U0 := Ω0 ∩ Ω1 .
Then {U0 , Uξ } forms an open covering of Ω1 . By choosing δ sufficiently small, we may assume that X ∩ B(ξ, δ) is a complex hypersurface. It follows from Theorem 3.15 (or Theorem 3.17) that there exists a bounded holomorphic function fξ on Ωξ such that fξ = f on X ∩ Ωξ . Since Ω0 is a pseudoconvex domain, there exists a holomorphic function f0 in Ω0 such that f0 = f on X ∩ Ω0 . Then f0 − fξ is holomorphic in Ωξ ∩ Ω0 and f0 − fξ = 0 on X ∩ Ωξ ∩ Ω0 . Since Uξ ∩ U0 ⊂ Ωξ ∩ Ω1 ∩ Ω0 , it follows from Theorem 5.22 that there exist f˜ξ ∈ Γ(Uξ , FV ) and f˜0 ∈ Γ(U0 , FV ) such that f0 − fξ = f˜0 − f˜ξ on Uξ ∩ U0 . We set g := f0 − f˜0 in U0 and g := fξ − f˜ξ in Uξ ∩ Ωξ . Then g is holomorphic in U0 ∪ (Uξ ∩ Ωξ ) and equals fξ − f˜ξ in Uξ ∩ Ωξ Therefore, g is bounded and holomorphic in Ω and satisfies g|V = f . If f ∈ A(V ), then we can prove similarly that there exists g ∈ A(Ω) such that g|V = f . More generally, Henkin-Leiterer [HER] proved bounded and continuous extensions in the case when Ω is a strictly pseudoconvex open
Integral Formulas for Strictly Pseudoconvex Domains
183
set (with not necessarily smooth boundary) in a Stein manifold without assuming that X intersects ∂Ω transversally. Amar [AMA2] also obtained bounded extensions of holomorphic functions from submanifolds of strictly pseudoconvex domains without assuming the transversality. Using the integral formula obtained by Hatziafratis, Hatziafratis [HAT2] proved the bounded extension of holomorphic functions from submanifolds in general position of strictly convex domains. Fornaess [FOR] investigated the integral formula by embedding strictly pseudoconvex domains into strictly convex domains. Adachi [ADA2; ADA3] proved bounded and continuous extensions from a submanifold V in general position of a weakly pseudoconvex domain Ω under the assumption that ∂V consists of strictly pseudoconvex boundary points of Ω. Using the method of Kerzman-Stein [KES] and the integral formula obtained by Hatziafratis [HAT1], Adachi-Kajimoto [ADK] obtained the holomorphic extension of Lipschitz functions from the boundary. Further, Jak´ obczak [JK1; JK2] studied extensions of holomorphic functions in various function spaces. 3.4
H p and C k Extensions
In this section we study H p (1 ≤ p < ∞) and C k (k = 1, 2, · · · , ∞) extensions of holomorphic functions from submanifolds in general position of a strictly pseudoconvex domain Ω in Cn with C ∞ boundary by following the methods of Beatrous [BEA] and Ahern-Schneider [AHS2], respectively. Let X be a closed submanifold in a neighborhood of Ω and let V = Ω ∩ X. Assume that X intersects ∂Ω transversally. We may assume that X = {zn = 0}. For f ∈ O(V ) ∩ Lp (V ) (1 ≤ p < ∞) and z ∈ Ω, we define (n − 1)! χ(ζ)(w(z, ζ))′ Ef (z) = ∧ ωζ ′ (ζ). f (ζ)ωζ ′ ζ) (2πi)n−1 V Φ(z,
By Theorem 3.13, Ef is holomorphic in Ω and Ef (z) = f (z) for z ∈ V . There exists a C ∞ (n − 1, n − 1) form η0 (z, ζ) on Ω × Ω with respect to ζ such that f (ζ)η0 (z, ζ) Ef (z) = . ζ)n Φ(z, V Then we obtain the following lemma.
Lemma 3.32 There exists a C ∞ (n − 1, n − 1) form η(z, ζ) on Ω × Ω with respect to ζ with the following properties:
Several Complex Variables and Integral Formulas
184
(a) η(·, ζ) is holomorphic in Ω for each ζ ∈ V fixed. (b) For f ∈ O(V ) ∩ L1 (V ) f (ζ)ρ(ζ)η(z, ζ) . Ef (z) = ζ)n+1 Φ(z, V Proof.
We choose a function ϕ ∈ C ∞ (R) with the following properties: 1 (|t| ≤ 1) 0 ≤ ϕ(t) ≤ 1 (t ∈ R), ϕ(t) = . 0 (|t| ≥ 2)
For ε > 0, we set ϕε (z) = ϕ
ρ(z) ε
.
Then ϕε (z) = 1 if |ρ(z)| ≤ ε and ϕε (z) = 0 if |ρ(z)| ≥ 2ε. We choose C ∞ N functions ψj , j = 1, · · · , N , in U2 with the properties that 1 = j=1 ψj (z) that for z ∈ U2 and there exists a constant c > 0 such if z ∈ supp(ψj ), then ∂ρ there exists a positive integer k = k(j) with ∂ζk (z) > c. Then we have Ef (z) =
N j=1
f (ζ)
V
Since
η0 (z, ζ)ψj (ζ) . ζ)n Φ(z,
¯ ∧ dζ¯k+1 ∧ · · · ∧ dζ¯n−1 = dζ¯1 ∧ · · · ∧ dζ¯k−1 ∧ ∂ρ
∂ρ ¯ dζ1 ∧ · · · ∧ dζ¯n−1 ∂ ζ¯k
on supp(ψi ), we obtain Ef (z) =
V
f (ζ)
¯ ∂ρ(ζ) ∧ ω(z, ζ) , ζ)n Φ(z,
where ω(z, ζ) is a C ∞ (n − 1, n − 2) form on Ω × V , and holomorphic with respect to z ∈ Ω for each fixed ζ ∈ V . Now we have ¯ ¯ ∂ρ(ζ) ∧ ω(z, ζ) ∂ρ(ζ)ϕ ε (ζ) ∧ ω(z, ζ) f (ζ) f (ζ) = n ζ) ζ)n Φ(z, Φ(z, V V ¯ ∂ρ(ζ)(1 − ϕε (ζ)) ∧ ω(z, ζ) f (ζ) + ζ)n Φ(z, V ε ε := I1 + I2 .
Integral Formulas for Strictly Pseudoconvex Domains
185
Then lim I1ε = 0. On the other hand, 1 − ϕε = 0 on ∂V , which implies that ε→0
f (ζ)ρ(ζ)(1 − ϕ ) ∧ ω(z, ζ) ε ∂¯ = ζ)n Φ(z, V ) ∧ ω(z, ζ) (1 − ϕ ε − f (ζ)ρ(ζ)∂¯ ζ)n Φ(z, V ¯ ε ∧ ω + Φ(1 ¯ − n(1 − ϕε )∂¯ζ Φ ∧ ω] − ϕε )∂ω ∂ϕ [−Φ =− f (ζ)ρ(ζ) . n+1 Φ V I2ε
We obtain
¯ ε ∧ ω f (ζ)ρ(ζ)Φ ∂ϕ V ρ(ζ) 1 ¯ Φ∂ρ ∧ ω = f (ζ)ρ(ζ)ϕ′ ε ε V |f (ζ)||ρ(ζ)| ≤C dVn−1 (ζ) ε X∩{−2ε≤ρ≤−ε} ≤C |f (ζ)|dVn−1 (ζ). X∩{−2ε≤ρ≤−ε}
Consequently, lim
ε→0
V
Thus we have Ef (z) = lim
ε→0
I2ε
¯ ε ∧ ω = 0. ∂ϕ f (ζ)ρ(ζ)Φ
=
We set
V
¯ − nω ∧ ∂¯ζ Φ) ∂ω f (ζ)ρ(ζ)(Φ . n+1 Φ
ζ)∂¯ζ ω(z, ζ) − nω(z, ζ) ∧ ∂¯ζ Φ(z, ζ). η(z, ζ) = Φ(z,
Then we have
Ef (z) =
V
Definition 3.22 X. Lemma 3.33
f (ζ)ρ(ζ)η(z, ζ) ζ)n+1 Φ(z,
(z ∈ Ω).
For z ∈ Ω, we denote by δX (z) the distance from z to
For 0 < ε < 1 we have
Several Complex Variables and Integral Formulas
186
(a) (b)
∂Ω
∂Ω
δX (z)−2ε dσ (z) ≤ Cε |ρ(ζ)|−ε . ζ)|n 2n−1 |Φ(z, dσ2n−1 (z) ≤ Cε |ρ(ζ)|−ε . n+ε |Φ(z, ζ)|
Proof. We set τ (z, ζ) = ImF (z, ζ), where F (z, ζ) is the Levi polynomial. We prove Lemma 3.33 in case n ≥ 3. We have
δX (z)−2ε dσ (z) ζ)n 2n−1 ∂Ω Φ(z, ∞ ≤C (|ρ(ζ)| + τ (z, ζ) + |w′ |2 + |w′′ |2 )−n |w′′ |−2ε Cn−2
C
0
×dτ dw′ dw′′ .
By the change of variables w′ =
τ (z, ζ) = |ρ(ζ)|x1 , we obtain
& |ρ(ζ)|x′ ,
w′′ =
& |ρ(ζ)|x′′ ,
δX (z)−2ε dσ (z) ζ)n 2n−1 ∂Ω Φ(z, ∞ & ≤C (1 + x1 + |x′ |2 + |x′′ |2 )−n |( |ρ(ζ)|)−2ε |x′′ |−2ε C
′
Cn−2 ′′
×dx1 dx dx
≤ C|ρ(ζ)|−ε −ε
≤ C|ρ(ζ)|
0
Cn−2
C∞ 0
∞
0
(1 + |x′ |2 + |x′′ |2 )−n+1 |x′′ |−2ε dx′ dx′′
(1 + r12 + r22 )−n+1 r12n−5 r2−2ε r2 dr1 dr2 .
Now we set r = λ cos θ, r2 = λ sin θ. Then ≤ C|ρ(ζ)|−ε
−ε
≤ Cε |ρ(ζ)|
−ε
≤ Cε |ρ(ζ)|
∞
0
0
1
π 2
(1 + λ2 )−n+1 λ2n−3−2ε (sin θ)1−2ε dλdθ
0
∞ ∞
(1 + λ2 )−n+1 λ2n−3−2ε dλ dλ ≤ Cε |ρ(ζ)|−ε . λ1+ε
Integral Formulas for Strictly Pseudoconvex Domains
187
This proves (a). By the same method as the proof of (a), we have
∂Ω
dσ2n−1 (z) ≤ Cδ(ζ)−ε ζ)|n+ε |Φ(z,
C
≤ Cε δ(ζ)−ε −ε
≤ Cε δ(ζ)
Cn−2 ∞
1
dx′ dx′′ (1 + |x′ |2 + |x′′ |2 )n−1+ε
dλ λ1+ε
.
This proves (b). Lemma 3.34
For 0 < ε < 1, we have
V
|ρ(ζ)|−ε dV (ζ) ≤ Cε (|ρ(z)| + δX (z)2 )−ε . ζ)|n n−1 |Φ(z,
Proof. We set δ(ζ) = |ρ(ζ)| and ζ ′ = (ζ1 , · · · , ζn−2 ). We may assume that (ρ(ζ), τ (z, ζ), ζ ′ ) forms a real coordinate system in a neighborhood of ∂Ω. Hence we have
δ(ζ)−ε dV (ζ) ζ)|n n−1 V |Φ(z, ∞ ∞ ≤C (δ(z) + δX (z)2 + δ(ζ) + τ (z, ζ) + |ζ ′ |2 )−n δ(ζ)−ε Cn−2
0
0
×dδdτ dζ ′ .
By a change of variables δ(ζ) = (δ(z) + δX (z)2 )x1 , τ (z, ζ) = (δ(z) + & 2 ′ δX (z) )x2 and ζ = (δ(z) + δX (z)2 x′ , we obtain
V
δ(ζ)−ε dV (ζ) ζ)|n n−1 |Φ(z, 2 −ε
≤ C(δ(z) + δX (z) ) ×dx1 dx2 dx′
2 −ε
≤ C(δ(z) + δX (z) ) ×dx1 dx′
2 −ε
≤ C(δ(z) + δX (z) )
Cn−2
0
Cn−2
0
∞
∞
0
∞
0
0
∞
∞
(1 + x1 + x2 + |x′ |2 )−n x−ε 1
(1 + x1 + |x′ |2 )−n+1 x−ε 1
2n−5 (1 + x1 + r2 )−n+1 x−ε dx1 dr. 1 r
188
Several Complex Variables and Integral Formulas
We set x1 = y12 . Then δ(ζ)−ε dV (ζ) ζ)|n n−1 V |Φ(z, ≤ C(δ(z) + δX (z)2 )−ε
0
∞
0
∞
(1 + y12 + r2 )−n+1 y11−2ε r2n−5 dy1 dr.
We set y1 = λ cos θ, r = λ sin θ. Then we obtain δ(ζ)−ε dV (ζ) ζ)|n n−1 V |Φ(z, π2 ∞ (1 + λ2 )−n+1 λ2n−3−2ε (cos θ)1−2ε dλdθ ≤ C(δ(z) + δX (z)2 )−ε 0 ∞ 0 2 −ε ≤ Cε (δ(z) + δX (z) ) (1 + λ2 )−n+1 λ2n−3−2ε dλ 0 ∞ dλ 2 −ε ≤ Cε (δ(z) + δX (z) ) 1+2ε λ 1 ≤ Cε (δ(z) + δX (z)2 )−ε .
Now we are going to prove H p extensions of Lp holomorphic functions in V . Theorem 3.20 Let Ω be a strictly pseudoconvex domain in Cn with smooth boundary and let X be a submanifold in a neighborhood of Ω which intersects ∂Ω transversally. Let V = X ∩ Ω and 1 ≤ p < ∞. If f ∈ Lp (V ), then Ef ∈ H p (Ω). Moreover, E : Lp (V ) → H p (Ω) is a continuous linear operator. Proof. We may assume that V = Ω ∩ X, where X = {zn = 0}. First we assume that 1 < p < ∞. Let q > 1 satisfy 1p + 1q = 1. We set δ(z) = |ρ(z)|. For any sufficiently small ε > 0 , by applying the H¨ older inequality and Lemma 3.34, we have |f (ζ)|δ(ζ)ε δ(ζ)−ε dVn−1 (ζ) |Ef (z)| ≤ C ζ)|n |Φ(z, V p1 1q δ(ζ)−εq |f (ζ)|p δ(ζ)εp dVn−1 (ζ) dV (ζ) ≤ ζ)|n ζ)|n n−1 |Φ(z, V |Φ(z, V p1 |f (ζ)|p δ(ζ)εp −2ε ≤ Cε δX (z) . dVn−1 (ζ) ζ)|n |Φ(z, V
Integral Formulas for Strictly Pseudoconvex Domains
189
Hence we obtain p
−2εp
|Ef (z)| ≤ Cε δX (z)
V
|f (ζ)|p δ(ζ)εp dVn−1 (ζ) . ζ)|n |Φ(z,
Using Fubini’s theorem and Lemma 3.33, we have |Ef (z)|p dσ2n−1 (z) ∂Ω −2εp δ (z) X ≤ Cε dσ (z) dVn−1 (ζ) |f (ζ)|p δ(ζ)εp ζ)|n 2n−1 ∂Ω |Φ(z, V ≤ Cε |f (ζ)|p δ(ζ)εp δ(ζ)−εp dVn−1 (ζ) V = Cε |f (ζ)|p dVn−1 (ζ). V
Thus Theorem 3.20 is proved in case 1 < p < ∞. Next we prove Theorem 3.20 for p = 1. By Lemma 3.32 we have f (ζ)ρ(ζ)η(z, ζ) dVn−1 (ζ), Ef (z) = ζ)n+1 Φ(z, V
where η(z, ζ) is a C ∞ function on Ω × Ω. Let 0 < ε < 1. Then we have |f (ζ)|δ(ζ)ε δ(ζ) |Ef (z)| ≤ C |f (ζ)| dVn−1 X(ζ) ≤ Cε dV (ζ). ζ)|n+1 ζ)|n+ε n−1 |Φ(z, V V |Φ(z, By Lemma 3.33, we obtain
∂Ω
|Ef (z)|dσ2n−1 (z) ≤ Cε ≤ Cε ≤ Cε
V
|f (ζ)|δ(ζ)ε
∂Ω
dσ2n−1 (z) ζ)|n+ε |Φ(z,
V
|f (ζ)|δ(ζ)ε δ(ζ)−ε dVn−1 (ζ)
V
|f (ζ)|dVn−1 (ζ).
dVn−1 (ζ)
Next we prove C k , k = 1, 2, · · · , ∞, extensions of holomorphic functions from submanifolds in general position of strictly pseudoconvex domains with C ∞ boundary. Lemma 3.35
Let Ω and V be the same notations as in Lemma 3.32.
190
Several Complex Variables and Integral Formulas
(a) For z ∈ Ω, there exists a constant C > 0 such that 1 dσ2n−3 (ζ) ≤ C| log |ρ(z)||. |Φ(z, ζ)|n−1 ∂V (b) For z ∈ Ω and δ > 0, there exists a constant C > 0 such that |ζ − z| dσ2n−3 (ζ) ≤ Cδ| log δ|. n−1 ∂V ∩B(z,δ) |Φ(z, ζ)| Proof. (a) Suppose δ > 0 is sufficiently small. Then we can choose a local coordinate system t = (t1 , · · · , t2n−3 ) on ∂V ∩ B(z, δ) such that t1 = Im Φ(z, ζ). We set t′ = (t2 , · · · , t2n−3 ). Then dt1 · · · dt2n−3 1 dσ (ζ) ≤ C 2n−3 n−1 2 n−1 |t|≤R (|ρ(z)| + t1 + |t| ) ∂V ∩B(z,δ) |Φ(z, ζ)| dt2 · · · dt2n−3 ≤C ′ 2 n−2 |t′ |≤R (|ρ(z)| + |t | ) R r2n−5 ≤C dr 2 n−2 0 (|ρ(z)| + r ) ≤ C| log |ρ(z)||. This proves (a). (b) By (a) we have 1 |ζ − z| dσ (ζ) ≤ Cδ dσ2n−3 (ζ) 2n−3 n−1 |Φ(z, ζ)| |Φ(z, ζ)|n−1 ∂V ∩B(z,δ) ∂V ∩B(z,δ) ≤ Cδ| log δ||.
Now we prove that every strictly pseudoconvex domain with C 2 boundary has a peak function by following Range [RAN2]. Lemma 3.36 Let Ω be a strictly pseudoconvex domain with C 2 boundary and let ζ ∈ ∂Ω. Then there exists a function f ∈ A(Ω) such that f (ζ) = 1 and |f (z)| < 1 for z ∈ Ω\{ζ}. Proof. There exists a neighborhood U of ∂Ω and a C 2 strictly plurisubharmonic function ρ in U such that Ω∩U = {z ∈ U | ρ(z) < 0}. Let F (z, ζ) be the Levi polynomial. It follows from (3.19) that there exists ε > 0 such that Re F (z, ζ) ≥ ρ(ζ) − ρ(z) + C|ζ − z|2
Integral Formulas for Strictly Pseudoconvex Domains
191
for ζ ∈ U , |z − ζ| ≤ ε. Choose ϕ ∈ C ∞ (Cn × Cn ) such that 0 ≤ ϕ ≤ 1 and ϕ(z, ζ) =
1 (|ζ − z| ≤ 2ε ) . 0 (|ζ − z| ≥ ε)
Fix ζ ∈ ∂Ω. Define λ(z) = ϕ(z, ζ)F (z, ζ) + (1 − ϕ(z, ζ))|ζ − z|2 . Then Re λ(z) > 0 for
z ∈ Ω\{ζ}.
Then there exists a neighborhood W of Ω\{ζ} such that Re λ(z) > 0 for ¯ = 0 on W ∩ B(ζ, ε/2), z ∈ W . We set u(z) = 1/λ(z) for z ∈ W . Since ∂u ∞ ¯ extends as a C (0,1)-form to a neighborhood of Ω. By Corollary 2.3 ∂u ¯ = ∂v ¯ in a neighborhood of there exists a function v ∈ C ∞ (Ω) such that ∂u −1 ¯ = 0 in W. Ω. Define g = (u − v + |v|Ω ) . Then Re g > 0 on Ω\{ζ} and ∂g −g Hence g is holomorphic in W . Define h = e . Since limz→ζ h(z) = 1, h is continuous on Ω and |h(z)| < 1 for z ∈ Ω\{ζ}. Definition 3.23 Let K(z, ζ) be the (2n − 3) form in Theorem 3.16. We write K(z, ζ) in the following form ζ)dσ2n−3 (ζ), K(z, ζ) = K(z,
: Ω × ∂V → C is a where dσ2n−3 is the surface measure on ∂V . Then K ∞ C function on Ω × ∂V and K(·, ζ) is holomorphic in Ω.
The following lemma is due to Ahern-Schneider [AHS1].
Lemma 3.37 Let Ω and V be the same notations as in Lemma 3.32. ζ) = 0 for all ζ ∈ ∂V . Then K(ζ, Proof.
0 , ζ0 ) = 0 for some ζ0 ∈ ∂V . We show that Assume that K(ζ f (ζ0 ) =
∂V
f (ζ)
0 , ζ) K(ζ dσ2n−3 (ζ) Φ(ζ0 , ζ)n−1
(3.33)
Several Complex Variables and Integral Formulas
192
for f ∈ A(V ). Let z ∈ V . By Theorem 3.16 we have
0 , ζ) K(ζ dσ2n−3 (ζ) Φ(ζ0 , ζ)n−1 ∂V
ζ) 0 , ζ) K(z, K(ζ = f (ζ) dσ2n−3 (ζ) − Φ(z, ζ)n−1 Φ(ζ0 , ζ)n−1 ∂V \B(ζ0 ,δ)
ζ) 0 , ζ) K(z, K(ζ + f (ζ) dσ2n−3 (ζ) − Φ(z, ζ)n−1 Φ(ζ0 , ζ)n−1 ∂V ∩B(ζ0 ,δ)
f (z) −
f (ζ)
:= J1 (z) + J2 (z).
It follows from Lebesgue’s dominated convergence theorem that limz→ζ0 J1 (z) = 0. On the other hand we have J2 (z) ≤
∂V ∩B(ζ0 ,δ)
+
|f (ζ)|
∂V ∩B(ζ0 ,δ)
0 , ζ)| |K(ζ dσ2n−3 (ζ) |Φ(ζ0 , ζ)|n−1
|f (ζ)|
:= J2′ (ζ0 ) + J2′ (z).
ζ)| |K(z, dσ2n−3 (ζ) |Φ(z, ζ)|n−1
By Lemma 3.35 (b) we have |J2 (ζ0 )| ≤ C|f |V δ| log δ|. To estimate J2′ (z), we let z approach ζ0 along the inward normal to ∂V . Then we have |z − ζ0 | ≤ C|z − ζ| and |ζ − ζ0 | ≤ C|z − ζ|. Consequently, ζ)| ≤ |K(z, ζ) − K(z, ζ0 )| + |K(z, ζ0 ) − K(ζ 0 , ζ0 )| |K(z, ≤ C(|ζ − ζ0 | + |z − ζ0 |) ≤ C|ζ − z|. If |z − ζ0 | < δ, then J2′ (z) ≤ C|f |V ≤ C|f |V
∂V ∩B(ζ0 ,δ)
∂V ∩B(z,2δ)
≤ C|f |V δ| log δ|.
|z − ζ| dσ2n−3 (ζ) |Φ(z, ζ)|n−1
|z − ζ| dσ2n−3 (ζ) |Φ(z, ζ)|n−1
By letting z approach ζ0 along the inward normal, we have (3.33). By Lemma 3.36 there exists f ∈ A(Ω) such that f (ζ0 ) = 1 and |f (z)| < 1 for
Integral Formulas for Strictly Pseudoconvex Domains
z ∈ Ω\{ζ0 }. Then f (ζ0 )N =
∂V
f (ζ)N
193
0 , ζ) K(ζ dσ2n−3 (ζ). Φ(ζ0 , ζ)n−1
By Lebesgue’s dominated convergence theorem, the right side of the above equality tends to 0 as N → ∞. This is a contradiction. Theorem 3.21 Let Ω be a strictly pseudoconvex domain with C ∞ boundary and let X = {zn = 0}, V = X ∩ Ω. Suppose X intersects ∂Ω transversally. If f ∈ O(V ) ∩ C k (V ), k = 0, 1, · · · , ∞, then E1 f ∈ O(Ω) ∩ C k (Ω). Proof.
We prove by induction on k that ζ) f (ζ)λ(z, ζ)K(z, G(z) = dσ2n−3 (ζ) n−1 Φ(z, ζ) ∂V
belongs to C k (Ω) if f ∈ O(V ) ∩ C k (V ) and λ ∈ C k+1 (Ω × ∂V ). Suppose k = 0. Then f (ζ)k(z, ζ) dσ2n−3 (ζ) G(z) = λ(z, z) n−1 ∂V Φ(z, ζ) f (ζ)k(z, ζ)(λ(z, ζ) − λ(z, z)) dσ2n−3 (ζ) + Φ(z, ζ)n−1 ∂V := G1 (z) + G2 (z). G1 is continuous on Ω by Theorem 3.18. On the other hand, we obtain ∂G2 λ1 (z, ζ)dσ2n−3 (ζ) λ2 (z, ζ)O(|z − ζ|)dσ2n−3 (ζ) (z) = + , n−1 ∂zj Φ(z, ζ) Φ(z, ζ)n ∂V ∂V where λ1 and λ2 are continuous functions on Ω×V . There exists a constant C > 0 such that & λ2 (z, ζ)O(|z − ζ|)dσ2n−3 (ζ) ≤ C |ρ(z)|| log |ρ(z)||. n Φ(z, ζ) ∂V
It follows from Lemma 3.20 that G2 ∈ Λα (Ω) for any 0 < α < 1/2. Hence G2 is continuous on Ω. Assume that the assertion has already been proved for k − 1. Let f ∈ O(V ) ∩ C k (V ) and λ ∈ C k+1 (Ω × ∂V ). If Ω = {z | ρ(z) < 0}, then dρ = 0 on ∂Ω. Let z 0 ∈ ∂V . We may assume that there exist constants σ1 > 0 and γ1 > 0 such that ∂ρ 0 ∂ζ1 (ζ) > γ1 for ζ ∈ B(z , σ1 ).
Several Complex Variables and Integral Formulas
194
In order to prove the assertion it is sufficient to show that G(z) =
∂V ∩B(z 0 ,σ1 )
ζ) f (ζ)λ(z, ζ)K(z, dσ2n−3 (ζ) Φ(z, ζ)n−1
belongs to C k (Ω). We may assume that (see Theorem 3.8) Φ(z, ζ) = F (z, ζ)M (z, ζ) for
(z, ζ) ∈ B(z 0 , σ1 ) × B(z 0 , σ1 ),
where F (z, ζ) is the Levi polynomial and M (z, ζ) = 0 for
((z, ζ) ∈ B(z 0 , σ1 ) × B(z 0 , σ1 ).
Then we obtain ∂Φ 0 0 ∂M 0 0 ∂F 0 0 (z , z ) = (z , z )F (z 0 , z 0 ) + M (z 0 , z 0 ) (z , z ) ∂ζ1 ∂ζ1 ∂ζ1 ∂ρ 0 = 2M (z 0 , z 0 ) (z ) = 0, ∂ζ1 ∂M ∂F ∂Φ 0 0 (z , z ) = ¯ (z 0 , z 0 )F (z 0 , z 0 ) + M (z 0 , z 0 ) ¯ (z 0 , z 0 ) = 0. ¯ ∂ ζ1 ∂ ζ1 ∂ ζ1 There exists γ2 > 0 such that −1 ∂Φ ∂ρ ∂Φ ∂ρ − > γ2 ∂ζ1 ∂ζ1 ∂ ζ¯1 ∂ ζ¯1
on B(z 0 , σ1 ) × B(z 0 , σ1 ).
We define
dζ = dζ1 ∧ · · · ∧ dζn−1 , [dζ]1 = dζ2 ∧ · · · ∧ dζn−1 ,
¯ 1 = dζ¯2 ∧ · · · ∧ dζ¯n−1 . [dζ]
Then we have d{Φ
−n−1
¯ 1 } = −mΦ−n [dζ]1 ∧ [dζ]
∂Φ ∂ρ − ∂ζ1 ∂ζ1
∂ρ ∂ ζ¯1
−1
∂Φ ∂ ζ¯1
¯1 dζ ∧ [dζ]
Integral Formulas for Strictly Pseudoconvex Domains
195
on ∂V ∩ B(z 0 , σ1 ). In view of Lemma 3.37 we may assume that K(z, ζ) = 0 for (z, ζ) ∈ B(z 0 , σ1 ) × B(z 0 , σ1 ). Then we obtain a representation ∂G (z) = ∂zj
∂V ∩B(z 0 ,σ1 )
+
ζ) f (ζ)λ1 (z, ζ)K(z, dσ2n−3 (ζ) Φ(z, ζ)n−1
∂V ∩B(z 0 ,σ1 )
1 (z) + G 2 (z), := G
ζ) f (ζ)λ2 (z, ζ)K(z, dσ2n−3 (ζ) n Φ(z, ζ)
where λ1 ∈ C k (Ω × ∂V ) and λ2 ∈ C k+1 (Ω × ∂V ). It follows from the induc 2 is expressed 1 ∈ C k−1 (Ω). On the other hand, G tive hypothesis that G by 2 (z) = G =
∂V ∩B(z 0 ,σ1 )
∂V ∩B(z 0 ,σ1 )
ζ) f (ζ)λ3 (z, ζ)K(z, ¯1 dζ ∧ [dζ] n Φ(z, ζ)
¯ 1 }. ζ)d{Φ−(n−1) [dζ]1 ∧ [dζ] f (ζ)λ4 (z, ζ)K(z,
Let ϕ(z, ζ) be a C ∞ function on Cn ×Cn satisfying ϕ = 0 on |z −ζ| > σ1 /2, ϕ = 1 on |z − ζ| < σ1 /4. Then 2 (z) G ¯ 1} ζ)ϕ(z, ζ)d{Φ−(n−1) [dζ]1 ∧ [dζ] = f (ζ)λ4 (z, ζ)K(z, ∂V ∩B(z 0 ,σ1 ) ¯ 1} ζ)(1 − ϕ(z, ζ))d{Φ−(n−1) [dζ]1 ∧ [dζ] + f (ζ)λ4 (z, ζ)K(z, ∂V ∩B(z 0 ,σ1 )
3 (z) + G 4 (z). := G
4 ∈ C k−1 (Ω). Using Stokes’ theorem, we have Clearly, G ¯ 1} ζ)ϕ(z, ζ)d{Φ−(n−1) [dζ]1 ∧ [dζ] G3 (z) = f (ζ)λ4 (z, ζ)K(z, ∂V ¯ 1. ζ)ϕ(z, ζ)}Φ−(n−1) [dζ]1 ∧ [dζ] d{f (ζ)λ4 (z, ζ)K(z, =− ∂V
3 ∈ C k−1 (Ω). Hence ∂ G ∈ It follows from the inductive hypothesis that G ∂zj ∈ C k (Ω). C k−1 (Ω). Similarly, we have ∂ G ∈ C k−1 (Ω), which means that G ∂ z¯j
Hence G ∈ C k (Ω). Thus we have proved that E1 f ∈ O(Ω) ∩ C k (Ω) if f ∈ O(V ) ∩ C k (V ).
196
Several Complex Variables and Integral Formulas
Theorem 3.20 was first proved by Cumenge [CUM]. Adachi [ADA1] and Elgueta [ELG] proved Theorem 3.21 in the case when k = ∞, independently. Jakobczak [JK1] also proved Theorem 3.21. The proof of Theorem 3.21 given here is an application of the method of Ahern-Schneider [AHS2]. Amar [AMA2] proved C ∞ extensions of holomorphic functions from submanifold of certain weakly pseudoconvex domains. In case 1 ≤ p < ∞, Adachi [ADA4] obtained Lp extensions of Lp holomorphic functions from submanifolds of strictly pseudoconvex domains with non-smooth boundary. Theorem 3.21 is still open when Ω is a strictly pseudoconvex domain with non-smooth boundary. 3.5
The Bergman Kernel
For the preparation of the next section, we study the Bergman kernel. We begin with an orthonormal system in a Hilbert space. Lemma 3.38 (Gram-Schmidt orthonormalization process) Suppose H is a Hilbert space. For a sequence {xn } of linearly independent vectors in H, we set x1 , e1 = x1 y2 y2 = x2 − (x2 , e1 )e1 , e2 = , y2 ··· n−1 yn , yn = xn − (xn , ek )ek , en = yn k=1 ··· . Then {en } is an orthonormal system. Proof. We prove Lemma 3.38 by induction on n. When n = 1, the proof is trivial. Assume that the assertion is true when n = m−1. Let 1 ≤ k < m. Then m−1 1 1 (xm , ej )(ej , ek ) (ym , ek ) = (em , ek ) = (xm , ek ) − ym ym j=1
1 = {(xm , ek ) − (xm , ek )(ek , ek )} = 0. ym
Since (em , em ) = 1, {e1 , · · · , en } is an orthonormal system.
Integral Formulas for Strictly Pseudoconvex Domains
197
Lemma 3.39 (Bessel’s inequality) Let H be a Hilbert space and let {x1 , · · · , xn } be an orthonormal system in H. Then n X
|(x, xk )|2 ≤ kxk2
k=1
for all x ∈ H. Proof. that
For any complex numbers α1 , · · · , αn , it follows from Lemma 3.39
n
2 n n
X
X X
αk xk = kαk xk k2 = |αk |2 .
k=1
Consequently,
2 n
X
αk xk =
x −
k=1
x−
k=1
n X
k=1
αk xk , x −
k=1 2
= kxk −
n X
αk xk
k=1
= kxk2 − = kxk2 −
k=1 n X
αk xk
k=1
x, n X
n X
!
αk (x, xk ) − |(x, xk )|2 +
k=1
−
!
n X
αk xk , x
+
αk (x, xk ) +
n X
k=1 n X
k=1 n X
!
n X
|αk |2
k=1
αk αk
k=1
|(x, xk ) − αk |2 .
k=1
If we set αk = (x, xk ), then 0 ≤ kxk2 −
n X
|(x, xk )|2 .
k=1
Lemma 3.40 Let H be a Hilbert space. Suppose {xn } is an orthonormal P∞ system in H and {αn } is a sequence of complex numbers. Then n=1 αn xn converges if and only if ∞ X
|αn |2 < ∞.
n=1
Proof.
For positive integers m, k with m > k > 0, we have
m
2 m
X
X
αn xn = |αn |2 .
n=k
n=k
(3.34)
Several Complex Variables and Integral Formulas
198
We set sm =
m
n=1
αn xn . Then 2
sm − sk =
m
n=k
|αn |2 .
(3.35)
If (3.34) holds, then by (3.35) {sm } is a Cauchy sequence, and hence {sn } converges. Conversely, if {sn } converges, then {sn } is a Cauchy sequence, and hence (3.34) follows from (3.35). Definition 3.24 Let H be a Hilbert space. An orthonormal system {xn } in H is said to be complete if ∞
x=
(x, xn )xn
n=1
for every x ∈ H. Lemma 3.41 Let H be a Hilbert space and let {xn } be an orthonormal system in H. Then {xn } is complete if and only if the following holds: (x, xn ) = 0 Proof.
(n = 1, 2, · · · )
=⇒
x = 0.
(3.36)
Let {xn } be complete. Then for any x ∈ H we have x=
∞
(x, xn )xn .
n=1
Hence (3.36) holds. Conversely, assume that (3.36) holds. It follows from the Bessel inequality that ∞
k=1
By Lemma 3.40, Then we have
∞
|(x, xk )|2 ≤ x2 .
n=1 (x, xn )xn
converges. We set y =
(x − y, xn ) = (x, xn ) − = (x, xn ) −
∞
k=1 ∞
(x, xk )xk , xn
(x, xk )(xk , xn )
k=1
= (x, xn ) − (x, xn ) = 0.
∞
n=1 (x, xn )xn .
Integral Formulas for Strictly Pseudoconvex Domains
199
Hence, by the assumption we have x − y = 0. Therefore we have x=
∞
(x, xn )xn .
n=1
Hence {xn } is complete.
Lemma 3.42 (Parseval’s equality) Let H be a Hilbert space and let {xn } be an orthonormal system in H. Then {xn } is complete if and only if x2 =
∞
n=1
|(x, xn )|2
(3.37)
for all x ∈ H. Proof.
Let x ∈ H. For a positive integer n we have
% %2 n n % % % % 2 (x, xk )xk % = x − |(x, xk )|2 . %x − % % k=1
k=1
Suppose {xn } is complete. Then the left side of the above equality converges to 0 as n → ∞. Hence (3.37) holds. Conversely, assume that (3.37) holds. Then we have x=
∞
(x, xn )xn ,
n=1
which implies that {xn } is complete.
Lemma 3.43 (Riesz-Fischer theorem) Let H be a Hilbert space and let {uj } be a complete orthonormal system in H. Then N (a) For x ∈ H, we set αj = (x, uj ). Then j=1 αj uj converges to x as ∞ N → ∞. Further, we have x2 = |αj |2 . j=1 ∞ (b) If j=1 |βj |2 < ∞, then there exists x ∈ H such that (x, uj ) = βj for all j, and 2
x =
∞ j=1
2
|βj | ,
x=
∞ j=1
βj u j .
Several Complex Variables and Integral Formulas
200
Proof.
We have already proved (a). Suppose xn =
n
∞
j=1
|βj |2 < ∞. We set
βj u j .
j=1
For n ≥ m > 0, we have %2 % % % n n % % 2 % % xn − xm = % βj u j % = |βj |2 → 0 % %j=m+1 j=m+1
(m, n → ∞).
Hence {xn } is a Cauchy sequence. Since H is a Hilbert space, {xn } converges. Let limn→∞ xn = x. Then we have x=
∞
βj u j .
j=1
Since xn 2 → x2 , we have x2 =
∞ j=1
|βj |2 .
This proves (b). Lemma 3.44
Define l 2 = α = {aj }
∞ 2 |a | < ∞, a ∈ C . j j j=1
For α = {aj }, β = {bj } ∈ l 2 , we define an inner product by (α, β) =
∞
aj b j .
j=1
Then l 2 is a Hilbert space. Further we have 1/2 ∞ = sup |(α, β)|. βl 2 = |bj |2 j=1
Proof.
α∈l 2
α≤1
For α ≤ 1, we have |(α, β)| ≤ αβ ≤ β.
Integral Formulas for Strictly Pseudoconvex Domains
201
On the other hand, if we set cj =
bj , kβk
γ = {cj },
then kγk = 1. Moreover we have |(γ, β)| = kβk. Lemma 3.45 are equivalent:
Let H be a Hilbert space. Then the following statements
(a) H is separable. (b) H contains a complete orthonormal system which is at most countable. Proof. (a) =⇒ (b). Suppose H is separable. Let E = {xn | n = 1, 2, · · · } ⊂ H, E = H. If xn is a linear combination of x1 , · · · , xn−1 , then we omit xn from E. Let {yn } be a subsequence of {xn } obtained by this process. Since {yn } is linearly independent, by the Schmidt orthonormalization process we have an orthonormal system {en }. The set of all linear combinations of elements in {en } is equal to the set of all linear combinations of elements in {xn }. Hence {en } is dense in H. Let x ∈ H. For any ε > 0, there exists positive integer N such that
N
X
cn xn < ε.
x −
n=1
Since
2 N N N
X X X
cn xn ≥ kx − (x, en )en k2 = kxk2 − |(x, en )|2 ,
x −
n=1
n=1
n=1
we have
kxk2 ≤
∞ X
|(x, en )|2 + ε2 .
n=1
Since ε > 0 is arbitrary, we have kxk2 ≤
∞ X
n=1
|(x, en )|2 .
Several Complex Variables and Integral Formulas
202
It follows from the Bessel inequality that 2
x =
∞
n=1
|(x, en )|2 .
Hence {en } is a complete orthonormal system. (b) =⇒ (a). Suppose H contains a complete orthonormal system {en } which is at most countable. We set k αn en | αn = an + ibn ∈ Q + iQ, k ∈ N . A= n=1
Then A is a countable set. For any x ∈ H we have x=
∞
(x, en )en .
n=1
Hence we have % % N % % % % (x, en )en % → 0 %x − % %
(N → ∞).
n=1
Then A = H, and hence H is separable.
Definition 3.25 Let Ω ⊂ Cn be an open set. We denote by A2 (Ω) the set of all holomorphic functions f in Ω satisfying |f (ζ)|2 dV (ζ) < ∞. Ω
A2 (Ω) is called the Bergman space. Lemma 3.46
A2 (Ω) is a closed subspace of L2 (Ω).
Proof. For simplicity, we prove Lemma 3.46 in case n = 1. The proof of the general case will be left to the reader. Let K ⊂ Ω be compact. We choose r > 0 such that B(w, r) ⊂ Ω for every w ∈ K. Let w ∈ K and h ∈ A2 (Ω). It follows from the Cauchy integral formula that h(w) =
1 2iπ
|z−w|=ρ
1 h(z) dz = z−w 2π
0
2π
h(w + ρeiθ )dθ
(3.38)
Integral Formulas for Strictly Pseudoconvex Domains
203
for 0 < ρ ≤ r. If we multiply by ρ and integrate from 0 to r, then we have r 2π 1 r2 iθ h(w) = h(w + ρe )dθ ρdρ 2 2π 0 0 1 = h(z)ρdρdθ 2π |z−w|≤r 1 = h(z)dxdy. 2π |z−w|≤r By the H¨older inequality we obtain 1 |h(z)|dxdy |h(w)| ≤ 2 πr |z−w|≤r 1/2 1/2 1 2 ≤ 2 |h(z)| dxdy dxdy πr |z−w|≤r |z−w|≤r 1/2 √ 1 πr |h(z)|2 dxdy ≤ 2 πr Ω 1 = √ h. πr Suppose fj ∈ A2 (Ω) for j = 1, 2, · · · , f ∈ L2 (Ω) and fj → f . For ε > 0, if we choose N sufficiently large, then for j, k ≥ N , we have √ fj − fk < ε πr. Therefore, we obtain 1 |fj (w) − fk (w)| ≤ √ fj − fk < ε πr for w ∈ K, which implies that {fj } converges to a holomorphic function f uniformly on every compact subset of Ω. Now we show that f ∈ A2 (Ω). Since {fj } is a Cauchy sequence, there exists positive number M such that fj ≤ M for all j. On the other hand, for any compact subset K ⊂ Ω |fj (z)|2 dxdy ≤ |fj (z)|2 dxdy = fj 2 ≤ M 2 . K
Ω
Hence we obtain
K
|f (z)|2 dxdy ≤ M 2 .
Several Complex Variables and Integral Formulas
204
Since K is independent of M , we have |f (z)|2 dxdy ≤ M 2 . Ω
Thus f ∈ A2 (Ω). Hence A2 (Ω) is a closed subspace of L2 (Ω). Theorem 3.22
A2 (Ω) has a countable complete orthonormal system.
Proof. Since A2 (Ω) is a separable Hilbert space (see Adams [ADM]), Theorem 3.22 follows from Lemma 3.45. Lemma 3.47 Then
Let {ϕj } be a complete orthonormal sequence in A2 (Ω). ∞ j=1
|ϕj (z)|2 = sup |f (z)|2 < ∞ f ∈A2 (Ω)
f ≤1
for all z ∈ Ω. Proof. Let K be a compact subset of Ω and let z ∈ K. By the RieszFischer theorem we have 2 ∞ ∞ 2 |ϕj (z)| = sup aj ϕj (z) = sup |f (z)|2 . {aj }∈l 2 f ∈A2 (Ω) j=1 j=1 {aj }l 2 ≤1
f ≤1
Consequently,
∞ j=1
|ϕj (z)|2 ≤ sup cK f 2 ≤ cK < ∞, f ∈A2 (Ω)
f ≤1
where cK is a constant depending only on K.
Lemma 3.48 Let Ω ⊂ Cn be an open set. For a ∈ Ω, define τa : A2 (Ω) → C by τa (f ) = f (a). Then τa is a bounded linear functional on A2 (Ω). Proof. It is clear that τa is a linear functional. We choose rj > 0 for j = 1, · · · , n such that {(z1 , · · · , zn ) | |zj − aj | ≤ rj , j = 1, · · · , n} ⊂ Ω.
Integral Formulas for Strictly Pseudoconvex Domains
205
It follows from the proof of Lemma 3.46 that 1 f (z1 , a2 , · · · , an )dx1 dy1 f (a) = 2 πr1 |z1 −a1 |≤r1 1 f (z1 , · · · , zn )dV. = n 2 · · · π r1 · · · rn2 |z1 −a1 |≤r1 |zn −an |≤rn It follows from the H¨ older inequality that there exists a constant Ca > 0 such that |τa (f )| = |f (a)| ≤ Ca f L2 . Thus τa is bounded.
Definition 3.26 By Lemma 3.48 and the Riesz representation theorem, there exists g ∈ A2 (Ω) such that τa (f ) = (f, g)
(f ∈ A2 (Ω)).
We define g(z) = KΩ (z, a). We say that KΩ : Ω × Ω → C is the Bergman kernel for Ω. By definition we obtain f (ζ)KΩ (ζ, z)dV (ζ) f (z) = Ω
Lemma 3.49
(f ∈ A2 (Ω)).
(3.39)
For any z, ζ ∈ Ω, KΩ (ζ, z) = KΩ (z, ζ).
Proof. For z ∈ Ω fixed, we have KΩ (·, z) ∈ A2 (Ω). If we set f (ζ) = KΩ (ζ, z), then (3.39) shows that KΩ (ζ, z) = f (ζ) = f (w)KΩ (w, ζ)dV (w) Ω = KΩ (w, z)KΩ (w, ζ)dV (w) Ω
=
KΩ (w, z)KΩ (w, ζ)dV (w)
Ω
= KΩ (z, ζ). It follows from Lemma 3.49 and (3.39) that f (z) = f (ζ)KΩ (z, ζ)dV (ζ) (f ∈ A2 (Ω), z ∈ Ω). Ω
(3.40)
Several Complex Variables and Integral Formulas
206
Lemma 3.50
There exists a constant C > 0 such that −n KΩ (·, a)L2 ≤ CδΩ (a)
(a ∈ Ω),
where δΩ (a) = dist(a, ∂Ω). √ Proof. We choose r such that r < δΩ (a)/ n. Then {z | |zi − ai | ≤ r} ⊂ Ω. Using the same method as the proof of Lemma 3.46, we have 1 f (z1 , · · · , zn )dV. ··· f (a) = n 2n π r |zn −an |≤r |z1 −a1 |≤r By H¨older’s inequality, f 2 |f (a)| ≤ √ nL n . ( π) r √ Letting r → δΩ (a)/ n, we have |f (a)| ≤
, n n (δΩ (a))−n f L2 . π
(3.41)
On the other hand, by the Riesz representation theorem we obtain τa = KΩ (·, a)L2 . It follows from (3.41) that , n n τa = sup |τa (f )| = sup |f (a)| ≤ (δΩ (a))−n . π f L2 =1 f L2 =1 We set , n n C= . π Then we have the desired inequality.
Lemma 3.51 Let K be a compact subset of Ω, Then there exists a constant CK > 0 such that for every complete orthonormal sequence {ϕj } in A2 (Ω), sup
∞
z∈K j=1
|ϕj (z)|2 ≤ CK .
Integral Formulas for Strictly Pseudoconvex Domains
207
Proof. For z ∈ K, we have δΩ (z) ≥ dist(K, ∂Ω). It follows from Lemma 3.50 that KΩ (·, z)L2 ≤ C(dist(K, ∂Ω)−n = CK
(z ∈ K).
Since KΩ (·, z) ∈ A2 (Ω), we have KΩ (ζ, z) =
∞
(KΩ (·, z), ϕj )ϕj (ζ).
(3.42)
j=1
By Lemma 3.42 we obtain ∞ j=1
|(KΩ (·, z), ϕj )|2 = KΩ (·, z)2L2 .
It follows from (3.40) that ϕj (ζ)KΩ (z, ζ)dV (ζ) = (ϕj , KΩ (·, z)). ϕj (z) =
(3.43)
Ω
Hence we have ∞ j=1
Theorem 3.23 Then
|ϕj (z)|2 = KΩ (·, z)2L2 ≤ CK .
Let {ϕj } be a complete orthonormal sequence in A2 (Ω).
KΩ (ζ, z) =
∞
ϕj (ζ)ϕj (z)
j=1
((ζ, z) ∈ Ω × Ω).
(3.44)
Moreover, the infinite series in the right side of (3.44) converges uniformly on every compact subset of Ω × Ω. Proof. If we substitute (3.43) into (3.42), then we obtain (3.44). Suppose K ⊂ Ω is compact. It is sufficient to show that the infinite series in the right side of (3.44) converges uniformly on K × K. It follows from Lemma 3.51 that 1/2 1/2 ∞ ∞ ∞ |ϕj (ζ)||ϕj (z)| ≤ |ϕj (ζ)|2 |ϕj (z)|2 ≤ CK j=1
j=1
j=1
Several Complex Variables and Integral Formulas
208
for ζ, z ∈ K. We set gn (z, ζ) =
n j=1
|ϕj (ζ)||ϕj (z)|.
Then {gn } converges monotonically on K. In view of Lemma 1.19 {gn } converges uniformly on K × K. Hence the infinite series in the right side of (3.44) converges uniformly on every compact subset of Ω × Ω. KΩ ∈ C ∞ (Ω × Ω).
Corollary 3.7
Proof. By Theorem 3.23, KΩ (ζ, z) is continuous in Ω×Ω. Since KΩ (ζ, z) is holomorphic with respect to (ζ, z¯), KΩ (ζ, z) is expressed by the Cauchy integral. Differentiating under the integral sign, derivatives of KΩ (ζ, z) of any order are continuous in Ω × Ω, which completes the proof of Corollary 3.7. Lemma 3.52 properties:
Ω : Ω × Ω → C satisfies the following Suppose a function K
Ω (z, ·) ∈ A2 (Ω) for every fixed z ∈ Ω. (1) K Ω (z, ζ)dV (ζ) for every f ∈ A2 (Ω). f (ζ)K (2) f (z) = Ω
Then
Proof.
Ω = KΩ . K
For z ∈ Ω, define kz (ζ) = KΩ (z, ζ). Since kz ∈ A2 (Ω), we obtain Ω (w, ζ)dV (ζ) kz (ζ)K KΩ (w, z) = KΩ (z, w) = kz (w) = =
Ω
Ω
=
Ω
Definition 3.27
Ω (w, ζ)dV (ζ) kz (ζ)K
Ω (w, ζ)KΩ (z, ζ)dV (ζ) K
Ω (w, z). =K
Let Ω be a domain in Cn . For a C 1 mapping F = (f1 , · · · , fn ) : Ω → Cn ,
Integral Formulas for Strictly Pseudoconvex Domains
209
define
∂f1 ∂z1 (z)
··· . .. F ′ (z) = .. . ∂fn ∂z1 (z) · · ·
∂f1 ∂zn (z) .. .
∂fn ∂zn (z)
.
Theorem 3.24 Let Ωj , j = 1, 2, be bounded domains in Cn and let F : Ω1 → Ω2 be a biholomorphic mapping. Then KΩ1 (ζ, z) = detF ′ (ζ)KΩ2 (F (ζ), F (z))detF ′ (z)
(ζ, z ∈ Ω1 ).
We set
Proof.
H(z, w) = detF ′ (ζ)KΩ2 (F (ζ), F (z))detF ′ (z). Then for a fixed point z ∈ Ω1 , H(z, ·) is holomorphic in Ω1 . Differentiating z = F −1 (F (z)), we have 1 = det(F −1 )′ (F (z))detF ′ (z). We set z˜ = F (z). Then using the Cauchy-Riemann equation, the Jacobian of F is equal to |detF ′ |2 . Hence we have
Ω1
=
|H(z, w)|2 dV (z)
Ω1
|detF ′ (z)|2 |KΩ2 (F (z), F (w))|2 |detF ′ (w)|2 dV (z)
z , F (w))|2 |detF ′ (w)|2 |det(F −1 )′ (˜ z )|2 dV |detF ′ (F −1 (˜ z ))|2 |KΩ2 (˜ = |detF ′ (w)|2 z , F (w))|2 |dV (˜ z ) < ∞. |KΩ2 (˜
=
Ω2
Ω2
Since H(z, w) = H(w, z), we have
Ω1
|H(z, w)|2 dV (w) < ∞,
which means that H(z, ·) ∈ A2 (Ω). Next we show that H(z, w) is the reproducing kernel for Ω1 . Let f ∈ A2 (Ω1 ). If we set ζ˜ = F (ζ), then we
Several Complex Variables and Integral Formulas
210
have
f (ζ)H(z, ζ)dV (ζ)
Ω1
f (ζ)detF ′ (z)KΩ2 (F (z), F (ζ))detF ′ (ζ)dV (ζ) Ω1 −1 ′ ˜ ˜ ˜ ˜ = detF ′ (z) (ζ). f (F −1 (ζ))det(F ) (ζ)KΩ (F (z), ζ)dV
=
2
Ω2
We set −1 ′ ˜ ˜ = f (F −1 (ζ))det(F ˜ g(ζ) ) (ζ).
Then g is holomorphic in Ω2 . Moreover, we have ˜ 2 dV (ζ), ˜ |g(ζ)| |f (ζ)|2 dV (ζ) = Ω2
Ω1
2
which implies that g ∈ A (Ω2 ). Thus we obtain ′ ˜ ˜ ˜ Ω (F (z), ζ)dV (ζ) g(ζ)K f (ζ)H(z, ζ)dV (ζ) = detF (z) 2 Ω2
Ω1
= detF ′ (z)g(F (z)) = f (z).
By Lemma 3.52, we have H(z, w) = KΩ1 (z, w) 3.6
Fefferman’s Mapping Theorem
We prove Fefferman’s mapping theorem [FEF] which says that every biholomorphic mapping between two strictly pseudoconvex domains with C ∞ boundary can be extended to a C ∞ mapping up to the boundary. BellLigocka [BEL] gave a simple proof of Fefferman’s mapping theorem. In what follows we give the proof of Fefferman’s mapping theorem by following the methods of Range [RAN2]. Range obtained C k extensions up to the boundary under the assumption that ∂Ω is of class C 2k+4 (1 ≤ k ≤ ∞). For simplicity, we assume that ∂Ω is of class C ∞ . In order to prove Fefferman’s mapping theorem we use the homotopy formula for strictly pseudoconvex domains constructed in 3.2. Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary. are as in 3.2. Suppose the neighborhood U of ∂Ω, the functions ϕ, w and Φ We will adopt the convention of denoting by C any positive constant which does not depend on the relevant parameters in the estimates.
Integral Formulas for Strictly Pseudoconvex Domains
211
If f is an L1 holomorphic function in Ω, then it follows from Theorem 3.12 that χ(ζ)w(z, ζ) n! ∧ ω(ζ) f (z) = LΩ f (z) = f (ζ)ωζ ζ) (2πi)n Ω Φ(z, for z ∈ Ω. Define
dV (ζ) = dx1 ∧ dx2 ∧ · · · ∧ dx2n for ζj = xj + ixj+n and j = 1, · · · , n. Then there exists a C ∞ function G(z, ζ) in Ω × Ω which is holomorphic in z ∈ Ω for fixed ζ ∈ Ω such that χ(ζ)w(z, ζ) n! ∧ ω(ζ) = G(z, ζ)dV (ζ). ωζ ζ) (2πi)n Φ(z,
Then we have
f (z) =
f (ζ)G(z, ζ)dV (ζ)
Ω
(z ∈ Ω).
(3.45)
For ζ ∈ U and −ε0 < ρ(ζ), we have χ(ζ) = 1. Hence for ζ ∈ U and −ε0 < ρ(ζ), we have n (z, ζ) w n! j ∧ ω(ζ). ∧ ∂¯ζ G(z, ζ)dV (ζ) = ζ) (2πi)n j=1 Φ(z,
If we choose ε > 0 (0 < ε < ε0 ) sufficiently small, then it follows from Theorem 3.8 and Theorem 3.9 that Φ(z, ζ) = F (z, ζ)M (z, ζ) =
n j=1
wj (z, ζ)(zj − ζj )
for |z − ζ| ≤ ε. Differentiating the above equality with respect to ζj we have ∂F (z, ζ)M (z, ζ) + O(|ζ − z|) ∂ζj ∂ρ(ζ) + O(|ζ − z|), = 2M (z, ζ) ∂ζj
wj (z, ζ) =
ζ) = (F (z, ζ) − 2ρ(ζ))M '(z, ζ), Φ(z,
Several Complex Variables and Integral Formulas
212
¯ ζ) = (∂¯ζ F (z, ζ) − 2∂ρ(ζ)) '(z, ζ). '(z, ζ) + (F (z, ζ) − 2ρ(ζ))∂¯ζ M M ∂¯ζ Φ(z,
LetV∂Ω be a neighborhood of ∂Ω. If we denote by N (z, ζ) any C ∞ form in V∂Ω × V∂Ω , then we have n wj ∂¯ζ Φ ∂¯ζ wj n! G(z, ζ)dV (ζ) = ∧ ω(ζ) + ∧ 2 (2πi)n j=1 Φ Φ =
= We set
n N wj ∧k=j ∂¯ζ wk ¯ n! ∧ ∂ζ Φ ∧ ω(ζ) + (−1)j−1 n n+1 (2πi) j=1 Φ Φn
¯ ' M n! ω ′ (w(z, ζ)) ∧ (∂¯ζ F − 2∂ρ(ζ)) N ∧ ω(ζ) + . n ζ)n+1 n (2πi) Φ(z, Φ
P = (P1 , · · · , Pn ) =
∂ρ ∂ρ ,··· , ∂ζ1 ∂ζn
.
Then we have G(z, ζ)dV (ζ) ¯ '(∂¯ζ F − 2∂ρ(ζ)) O(|z − ζ|) N n! ω ′ (P (ζ))(2M )n M ∧ ω(ζ) + + n n+1 n+1 n (2πi) Φ Φ Φ ′ n n ¯ O(|z − ζ|) n! ω (P (ζ)) ∧ (−2∂ρ(ζ))2 M N ∧ ω(ζ) + = + . n+1 n 'n (2πi)n (F (z, ζ) − 2ρ(ζ))n+1 M Φ Φ =
On the other hand, for ζ0 ∈ ∂Ω with |ζ − ζ0 | = dist(ζ, ∂Ω), we have '(z, ζ0 ). Hence we have M (z, ζ0 ) = M M (ζ, ζ)n M (z, ζ)n = + O(|ζ − z|) = 1 + O(|ζ − ζ0 |) + O(|ζ − z|). '(z, ζ)n '(ζ, ζ)n M M
Consequently, for ζ ∈ Ω we obtain
Further we have
M (z, ζ)n 1 O(|ζ − z|) = + . n ζ) ' Φ(z, ζ) Φ(z, M (z, ζ) Φ(z, ζ)
n−1 ¯ ¯ ¯ ∂ρ(ζ) ∧ ω ′ (P (ζ)) ∧ ω(ζ) = ∂ρ(ζ) ∧ ∂ρ(ζ) ∧ (∂∂ρ(ζ)) .
We set n−1 ¯ ¯ ∧ ∂ρ(ζ) ∧ (∂∂ρ(ζ)) . H(ζ) = (2πi)−n ∂ρ(ζ)
Integral Formulas for Strictly Pseudoconvex Domains
213
Then we have H(ζ) = H(ζ). Therefore, if we write H(ζ) = h(ζ)dV (ζ), then h is a real-valued function. Hence we have the following lemma. Lemma 3.53 For (z, ζ) ∈ Ω × Ω with −ε < ρ(ζ) and |z − ζ| < ε, there exists a real-valued function a(ζ) such that G(z, ζ) = Lemma 3.54
a(ζ) N (z, ζ) O(|ζ − z|) + . + n+1 (F (z, ζ) − 2ρ(ζ))n+1 Φ(z, ζ) Φ(z, ζ)n
Let δΩ (z) = dist(z, ∂Ω). For z ∈ Ω, define Iα :=
Ω
dV (ζ) . |Φ(z, ζ)|n+1+α
Then there exists a constant C > 0 such that (a) If α < 0, then Iα < C. (b) If α = 0, then Iα ≤ C| log δΩ (z)|. (c) If α > 0, then Iα ≤ C(δΩ (z))−α .
Proof. We may assume that ρ(ζ) > −ε and |z − ζ| ≤ ε. There exists a constant β > 0 such that ζ) ≥ −ρ(ζ) − ρ(z) + β|ζ − z|2 . Re Φ(z,
ζ), We choose a coordinate system t = (t1 , · · · , t2n ) such that t1 = Im Φ(z, t2 = ρ(ζ). We set t′ = (t3 , · · · , t2n ). Then we have Iα ≤ C ≤C ≤C
|t|≤M
|t′ |≤M
0
M
dt (|t1 | + |t2 | + |ρ(z)| + |t|2 )n+1+α dt′ (|ρ(z)| + |t′ |2 )n−1+α
r2n−3 dr. (|ρ(z)| + r2 )n−1+α
In case α < 0, we have Iα ≤ C
0
M
1 r1+α
dr ≤ C.
Several Complex Variables and Integral Formulas
214
& In case α ≥ 0, if we set r = λ |ρ(z)|, then M/√|ρ(z)| λ2n−3 Iα ≤ C dλ α |ρ(z)| (1 + λ2 )n−1+α 0 M/√|ρ(z)| C dλ ≤ . |ρ(z)|α 1 λ1+2α In case α = 0, we have I0 ≤ C
M/
√
|ρ(z)|
1
dλ ≤ C| log |ρ(z)|| ≤ C| log δΩ (z)|. λ
In case α > 0, we have Iα ≤ Definition 3.28
C |ρ(z)|α
1
∞
dλ ≤ C(δΩ (z))−α . λ1+2α
For z, ζ ∈ Ω, we define
F (z, ζ) = F (z, ζ) − 2ρ(ζ),
Lemma 3.55
We have
F ∗ (z, ζ) = F (ζ, z).
(a) F (z, ζ) − F∗ (z, ζ) = O(|ζ − z|3 ). (b) If ζ, z ∈ Ω are sufficiently close to ∂Ω, then we have |F ∗ | ≥ C|F |. Proof.
(a) It follows from Taylor’s formula that n
∂2ρ ∂ρ ∂ρ (ζ) = (z) + (z)(ζk − zk ) ∂ζj ∂ζj ∂ζj ∂ζk k=1
+
n ∂2ρ (z)(ζ¯k − z¯k ) + O(|ζ − z|2 ), ¯ ∂ζ ∂ ζ j k k=1
(3.46)
∂2ρ ∂2ρ (ζ) = (z) + O(|ζ − z|), ∂ζj ∂ζk ∂ζj ∂ζk
(3.47)
∂2ρ ∂2ρ (z) = (ζ) + O(|ζ − z|). ∂ζj ∂ ζ¯k ∂ζj ∂ ζ¯k
(3.48)
Integral Formulas for Strictly Pseudoconvex Domains
215
By definition, we have F (z, ζ) = 2
n n ∂ 2ρ ∂ρ (ζ)(ζj − zj ) − (ζ)(ζj − zj )(ζk − zk ). (3.49) ∂ζj ∂ζj ∂ζk j=1 j,k=1
Substituting (3,46), (3.47) and (3.48) into (3.49), we obtain F (z, ζ) = −F (ζ, z)+2
n
∂2ρ 3 ¯ ¯k (ζ)(ζj −zj )(ζk − z¯k )+O(|ζ −z| ). (3.50) ∂ζ ∂ ζ j j,k=1
On the other hand, it follows from Taylor’s formula that 1 1 ρ(z) = ρ(ζ) − F (z, ζ) − F (z, ζ) 2 2 n 2 ∂ ρ + (ζ)(ζj − zj )(ζ¯k − z¯k ) + O(|ζ − z|3 ). ∂ζj ∂ ζ¯k j,k=1
Substituting (3.50) into the above equality we obtain 1 1 ρ(z) = ρ(ζ) − F (z, ζ) + F (ζ, z) + O(|ζ − z|3 ). 2 2 This proves (a). (b) We may assume that |z − ζ| is sufficiently small. Then there exist positive constants C1 and C2 such that |F ∗ | ≥ |F | − |F ∗ − F | ≥
This proves (b).
Definition 3.29
1 1 |F | + C1 |ζ − z|2 − C2 |ζ − z|3 ≥ |F|. 2 2
For (z, ζ) ∈ Ω × Ω, define G∗ (z, ζ) := G(ζ, z)
and B(z, ζ) := G(z, ζ) − G∗ (z, ζ). By definition, we have B(ζ, z) = −B(z, ζ). Theorem 3.25 C > 0 such that
Let s < (2n + 2)/(2n + 1). Then there exists a constant
Ω
|B(z, ζ)|s dV (ζ) < C
(z ∈ Ω)
Several Complex Variables and Integral Formulas
216
and
Ω
Proof.
|B(z, ζ)|s dV (z) < C
(ζ ∈ Ω).
Define C(z, ζ) :=
If we prove the inequality
a(ζ) a(z) − . n+1 ∗ F (z, ζ) F (z, ζ)n+1
|C(z, ζ)| ≤ C
1 ζ)|n+ 2 |Φ(z,
1
,
(3.51)
ζ)|n+(1/2) . Therethen it follows from Lemma 3.53 that |B(z, ζ)| ≤ |Φ(z, fore, it is sufficient to show (3.51). Since a(ζ) is a real-valued C ∞ function, we obtain a(ζ) a(z) 1 1 − = a(ζ) − F(z, ζ)n+1 F ∗ (z, ζ)n+1 F (z, ζ)n+1 F∗ (z, ζ)n+1 +
Since
O(|ζ − z|) . F ∗ (z, ζ)n+1
|F(z, ζ)| ≥ C|ζ − z|2
(ζ, z ∈ Ω),
it follows from Lemma 3.55 that 1 1 − F (z, ζ)n+1 F∗ (z, ζ)n+1 n 1 1 ∗ = (F − F ) n+1−ν ν+1 (z, ζ) F ∗ (z, ζ) ν=0 F 1 1 ≤ C|ζ − z|3 ≤C |F(z, ζ)|n+2 |F (z, ζ)|n+1/2 ≤C
1
ζ)|n+1/2 |Φ(z,
.
This proves (3.51). Thus we obtain ζ)|−(2n+1)s/2 . |B(z, ζ)|s ≤ C|Φ(z,
Integral Formulas for Strictly Pseudoconvex Domains
217
By Lemma 3.55, |B(z, ζ)|s is integrable, provided s < (2n + 2)/(2n + 1). The second inequality follows from the equality B(z, ζ) = −B(ζ, z). Definition 3.30 A C ∞ function A(z, ζ) in Ω × Ω is called a simple admissible kernel of order λ = 2n + j − 2t + 2 if for any P ∈ ∂Ω, there exist a neighborhood U of P and C ∞ functions Ej (z, ζ) in U × U such that A(z, ζ) has a representation A(z, ζ) =
Ej (z, ζ) ζ)t Φ(z,
or A(z, ζ) =
Ej (z, ζ) , Φ∗ (z, ζ)t
where j and t are positive integers with t ≥ 2, and Ej (z, ζ) satisfy the inequalities |Ej (z, ζ)| ≤ C|ζ − z|j
((z, ζ) ∈ U × U ).
Definition 3.31 A C ∞ function A(z, ζ) in Ω × Ω is called an admissible kernel of order λ if for any positive integer N there exist simple admissible kernels A(0) , · · · , A(N −1) of order ≥ λ such that A=
N −1 s=0
A(s) + R(N ) ,
where R(N ) satisfies that for any nonnegative integer k, if we choose N sufficiently large, then f (ζ)R(N ) (·, ζ)dV (ζ) Ω
Lemma 3.56 Then
(f ∈ L2 (Ω)).
We denote by Aλ a simple admissible kernel of order λ.
Ω
Proof.
k,Ω
≤ Ck f L2
|Aλ (z, ζ)|dV (ζ) ≤ C
1 (λ > 0) | log δΩ (z)| (λ = 0) . (δΩ (z))λ/2 (λ < 0)
We choose a local coordinate system u1 , · · · , u2n such that ρ(ζ) = u1 ,
ζ) = u2 , Im Φ(z,
|u| ≈ |ζ − z|.
Several Complex Variables and Integral Formulas
218
We set u = (u1 , · · · , u2n ), u′ |A(z, ζ)|dV (ζ) = Ω
≤ ≤ ≤ ≤
= (u3 , · · · , u2n ). Then we have E (z, ζ) j dV (ζ) ζ)t Ω Φ(z, |u|j du C 2 t |u|≤M (|u1 | + |u2 | + |ρ(z)| + |u| ) du C 2 t−j/2 |u|≤M (|u1 | + |u2 | + |ρ(z)| + |u| ) du′ ′ 2 t−2−j/2 |u′ |≤M (|ρ(z)| + |u | ) M r2n−3 C dr. (|ρ(z)| + r2 )t−2−j/2 0
In case λ = 2n + j − 2t + 2 > 0, we have M r2n−2t+j+1 dr = C |A(z, ζ)|dV (ζ) ≤ C
0
0
Ω
M
rλ−1 dr ≤ C.
& In case λ = 2n + j − 2t + 2 ≤ 0, if we set r = |ρ(z)|s, then M/√|ρ(z)| s2n−3 λ/2 ds |A(z, ζ)|dV (ζ) ≤ C|ρ(z)| (1 + s2 )t−2−(j/2) 0 Ω M/√|ρ(z)| ds λ/2 . ≤ C|ρ(z)| 1−λ s 1
Lemma 3.57 Let Ej (z, ζ) be a C ∞ function on Ω × Ω such that |Ej (z, ζ)| ≤ |ζ − z|j . For positive integers t1 , t2 , we set A(z, ζ) =
Ej (z, ζ) . F (z, ζ)t1 F ∗ (z, ζ)t2
(3.52)
Then A(z, ζ) is an admissible kernel of order λ = 2n + j − 2(t1 + t2 ) + 2. Proof.
We have t 2 −1 1 1 1 . = + (F − F ∗ ) t −ν ∗ t t 2 2 2 (F∗ )ν+1 (F ) (F ) ν=0 F
(3.53)
Substituting (3.51) into (3.50), we obtain A=
t2 −1 1 Ej ∗ Ej . + (F − F ) t t t t −ν 2 F 1F 2 F 1 (F ∗ )ν+1 ν=0 F
(3.54)
Integral Formulas for Strictly Pseudoconvex Domains
219
If we replace t2 by ν + 1 in (3.53) and substitute it into the right side of (3.54), we obtain A=
Ej (F − F ∗ ) Ej + F t1 Ft2 Ft1 +t2 +1 t2 −1 1 Ej (F − F ∗ )2 . + t t +1−µ 1 2 F (F∗ )µ+1 µ=0 F
Repeating this process, we have A=
N −1 s=0
t 2 −1 Ej (F − F ∗ )s Ej + (F − F ∗ )N . (3.55) t +t +s t +t +N −1−ν 1 2 1 2 F F (F∗ )ν+1 ν=0
Each term of the first sum in the right side of (3.55) is a simple admissible kernel of degree ≥ λ. We denote the second sum in the right side of (3.55) by R(N ) . It follows from Lemma 3.55 that (F − F ∗ )N Ej = O(|ζ − z|3N +j ).
Hence the absolute values of k-th order derivatives of R(N ) are bounded by the sum of C Since
|ζ − z|j+3N −µ
|F (z, ζ)|t1 +t2 +N +k−µ
|ζ − z|j+3N −µ
|F (z, ζ)|t1 +t2 +N +k−µ
(0 ≤ µ ≤ k).
(3.56)
≤ C|ζ − z|j+N −2(t1 +t2 +k)
for N with N ≥ 2(t1 +t2 +k), derivatives of R(N ) of order ≤ k are bounded. Thus A(z, ζ) is an admissible kernel of order λ. Theorem 3.26 order 1. Proof. that
B(z, ζ) = G(z, ζ) − G∗ (z, ζ) is an admissible kernel of
By Lemma 3.53, there exists a real-valued C ∞ function a(ζ) such G(z, ζ) =
a(ζ) O(|ζ − z|) N (z, ζ) + + . n+1 n+1 F (z, ζ) Φ(z, ζ) Φ(z, ζ)n
Thus we have a representation B(z, ζ) = a(ζ)
1
F(z, ζ)n+1
−
1 F ∗ (z, ζ)n+1
+ A1 ,
Several Complex Variables and Integral Formulas
220
where A1 is an admissible kernel of order 1. Since 1 Fn+1
−
1 (F ∗ )n+1
= (F∗ − F)
n
1 , n+1−ν (F ∗ )ν+1 ν=0 F
by Lemma 3.57 1/F n+1 − 1/(F ∗ )n+1 is an admissible kernel of order 2n + 3 − 2(n + 2) + 2 = 1.
Definition 3.32 A vector field L on Ω is said to be a tangent vector field for ∂Ω if Lρ = 0 on ∂Ω for any defining function ρ for Ω. Lemma 3.58 Let Ω ⊂ Rn be a bounded domain with C ∞ boundary and let L be a tangent vector field of class C ∞ for ∂Ω. Then there exists a first order partial differential operator L∗ on Ω of class C ∞ such that (f, Lg)Ω = (L∗ f, g)Ω for all f, g ∈ C ∞ (Ω). Proof. Assume that there exists an open set U such that supp(L) ⊂⊂ U , U ∩ ∂Ω = φ and such that if we set ρ(ζ) = x1 , then x1 , · · · , xn form a coordinate system in U . We set x′ = (x2 , · · · , xn ), and L=
n
aj
j=1
∂ ∂xj
(aj ∈ Cc∞ (U )).
Then we have Lr(0, x′ ) = a1 (0, x′ ) = 0. The volume element dV = γ(x)dx1 ∧ · · · ∧ dxn satisfies γ(x) > 0 for x ∈ U . If f, g ∈ C ∞ (U ), then we have n ∂¯ g (f, Lg)Ω∩U = fa ¯j γ(x)dx1 · · · dxn ∂x j j=1 x∈U,x1 ≤0 n ∂ (f a ¯j γ)¯ gdx1 · · · dxn =− ∂x j j=1 x∈U,x1 ≤0 fa ¯1 γ¯ gdx2 · · · dxn + x∈U,x1 =0
n
∂ (f a ¯j γ)¯ gdx1 · · · dxn ∂xj j=1 x∈U,x1 ≤0
n ∂f −1 ∂ +γ (¯ aj γ)f g¯dV (x). =− a ¯j ∂xj ∂xj j=1 x∈U,x1 ≤0
=−
Integral Formulas for Strictly Pseudoconvex Domains
221
Hence if we define L∗ := −
n j=1
a ¯j
n ∂ ∂ − γ −1 (¯ aj γ), ∂xj ∂xj j=1
then we have (f, Lg)Ω = (f, Lg)Ω∩U =
L∗ f g¯dV (ζ) = (L∗ f, g)Ω .
Ω∩U
In the general case, we can prove Lemma 3.58 using a partition of unity argument. Lemma 3.59 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let ρ be a defining function for Ω. Then the vector field Y =
n n ∂ρ ∂ ∂ρ ∂ − ¯ ∂ ζj ∂ζj j=1 ∂ζj ∂ ζ¯j j=1
is a tangent vector field for ∂Ω of class C ∞ and satisfies (Y Φ)(ζ, ζ) = 0 for ζ ∈ ∂Ω.
Proof. Let ρ˜ be a defining function for Ω. Then there exist a C ∞ function γ(z) > 0 such that ρ˜(ζ) = γ(ζ)ρ(ζ) (see Lemma 1.21). Hence we have Y ρ(ζ) = 0 for ζ ∈ ∂Ω. Thus Y is a tangent vector field for ∂Ω. We obtain for ζ ∈ ∂Ω ')(ζ, ζ) = Y {(F − 2ρ)M '}(ζ, ζ) = (Y F )(ζ, ζ)M '(ζ, ζ). (Y Φ)(ζ, ζ) = Y (F M
'(ζ, ζ) = 0 for ζ ∈ ∂Ω and But we have M 2 n ∂ρ Y F (ζ, ζ) = 2 ∂ζj (ζ) = 0 j=1 which means that (Y Φ)(ζ, ζ) = 0 for ζ ∈ ∂Ω.
(ζ ∈ ∂Ω),
Lemma 3.60 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary and let Aλ denote an arbitrary admissible kernel of order λ, where λ is equal to 0 or 1. Suppose V (z) is a vector field with respect to z of class C ∞ on Ω. Then V (z) f (ζ)Aλ (z, ζ)dV (ζ) = (Y ∗ f )(ζ)Aλ (z, ζ)dV (ζ) Ω Ω f (ζ)Aλ (z, ζ)dV (ζ). + Ω
222
Several Complex Variables and Integral Formulas
Proof. We set ∆∂Ω = {(ζ, ζ) ∈ C2n | ζ ∈ ∂Ω}. Let W be a neighborhood ζ) = 0 for (z, ζ) ∈ W . We choose ψ ∈ Cc∞ (W ) of ∆∂Ω such that (Y Φ)(z, with the properties that 0 ≤ ψ ≤ 1, and ψ = 1 in a neighborhood W ′ (⊂⊂ W ) of ∆∂Ω . ψAλ is expressed by ψAλ =
Ej , t Φ
where λ = 2n + j − 2t + 2. Then we have
Ej 1 1 Ej 1 Y Ej 1 ψAλ = ψ Y =ψ − +ψ t t−1 t−1 Y Φ t−1 t−1Φ Φ Φ YΦ Ej Ej−1 = ψ1 Y + t−1 t−1 Φ Φ = ψ1 Y Aλ+2 + ψ2 Aλ+1 ,
where ψ1 and ψ2 are C ∞ functions with compact supports in W . Consequently, f (ζ)Aλ (z, ζ)dV (ζ) = f (ζ)V (z) Aλ (z, ζ)dV (ζ) V (z) Ω Ω = f (ζ)Aλ−2 (z, ζ)dV (ζ) Ω f {ψAλ−2 + (1 − ψ)Aλ−2 } dV (ζ). = Ω
∞
(1 − ψ)Aλ−2 is of class C on Ω × Ω. On the other hand, we have f (ψ1 Y Aλ + ψ2 Aλ−1 )dV (ζ) f ψAλ−2 dV (ζ) = Ω Ω = Y ∗ (f ψ1 )Aλ dV (ζ) Ω f (ψ3 Y Aλ+1 + ψ4 Aλ )dV (ζ) + Ω f Aλ dV (ζ), = (Y ∗ f )Aλ dV (ζ) + Ω
Ω
which completes the proof of Lemma 3.60. Lemma 3.61
Define Aλ f (z) =
Then
Ω
f (ζ)Aλ (z, ζ)dV (ζ).
Integral Formulas for Strictly Pseudoconvex Domains
223
(a) A0 is a bounded operator from Λα (Ω) to Λα/2 (Ω) for every 0 < α < 1. (b) A1 is a bounded operator from L∞ (Ω) to Λ1/2 (Ω). Proof. First we prove (b). It is sufficient to prove the inequality (see Lemma 3.20) |dz A1 f (z)| ≤ C|f |Ω δΩ (z)−1/2
(z ∈ Ω).
(3.57)
By Lemma 3.56, we have |dz A1 f (z)| = f (ζ)A−1 (z, ζ)dV (ζ) Ω |A−1 (z, ζ)|dV (ζ) ≤ |f |Ω Ω
≤ CδΩ (z)−1/2 .
This proves (3.57). Next we prove (a). It is sufficient to show that |dz (A0 f )(z)| ≤ C|f |α,Ω δΩ (z)−1+α/2
(z ∈ Ω).
(3.58)
∂ ∂zj
or ∂∂z¯j . Then we have (z) f (ζ)A−2 (z, ζ)dV (ζ) V (A0 f ) = Ω = (f (ζ) − f (z))A−2 (z, ζ)dV (ζ) Ω +f (z) A−2 (z, ζ)dV (ζ).
Let V (z) denote either
Ω
≥ C|ζ − z|2 , it follows from Lemma 3.54 and the definition of the Since |Φ| degree of the admissible kernel that |ζ − z|α+j dV (ζ) |(f (ζ) − f (z))A−2 (z, ζ)|dV (ζ) ≤ C|f |α n+2+(j/2) |Φ| Ω Ω −(n+1+1−(α/2)) dV (ζ) ≤ C|f |α |Φ| Ω
≤ C|f |α δΩ (z)−1+α/2 .
On the other hand, using the same method as the proof of Lemma 3.60, we obtain ψA−2 = ψ1 Y A0 + ψ1 A−1 .
Several Complex Variables and Integral Formulas
224
Hence we have Z Z Z ∗ ψA−2 (z, ζ)dV (ζ) = Y ψ1 A0 dV (ζ) + ψ1 A−1 dV (ζ) Ω Ω ZΩ A−1 dV (ζ). = Ω
By Lemma 3.56, we obtain Z Z ψA−2 (z, ζ)dV (ζ) ≤ |A−1 | dV (ζ) ≤ δΩ (z)−1/2 ≤ CδΩ (z)−1+(α/2) , Ω
Ω
which completes the proof of Lemma 3.61.
Definition 3.33 For multi-indices α = (α1 , · · · , αn ) and β (β1 , · · · , βn ), where αj , βj are nonnegative integers, define
=
∂ |α|+|β|
¯
∂zαβ = Definition 3.34
∂z1α1
· · · ∂znαn ∂ z¯1β1 · · · ∂ z¯nβn
.
For f ∈ L1 (Ω), define
Gf (z) :=
Z
f (ζ)G(z, ζ)dV (ζ)
Z
f (ζ)G∗ (z, ζ)dV (ζ)
Z
f (ζ)B(z, ζ)dV (ζ)
(z ∈ Ω),
Ω
∗
G f (z) :=
(z ∈ Ω),
Ω
Bf (z) :=
(z ∈ Ω).
Ω
Theorem 3.27 Let k be a nonnegative integer. Then operators G, G∗ and B have the following properties: (a) G and G∗ are bounded operators from C k+α (Ω) to C k+(α/2) (Ω) for every 0 < α < 1. (b) B is a bounded operator from C k (Ω) to C k+(1/2) (Ω). Proof.
Let |α| + |β| = j ≤ k. Since G(z, ζ) is an admissible kernel of
Integral Formulas for Strictly Pseudoconvex Domains
225
order 0, it follows from Lemma 3.60 that ¯ αβ¯ f (ζ)∂zαβ G(z, ζ)dV (ζ) ∂z (Gf )(z) = Ω ¯ f (ζ)∂zαβ A0 (z, ζ)dV (ζ) = Ω
=
j ν=0
By Lemma 3.61, we obtain
(ν)
Ω
((Y ∗ )ν f )A0 (z, ζ)dV (ζ).
¯
|∂ αβ (Gf |α ≤ C
j
ν=0
|(Y ∗ )ν f |α/2 .
Therefore, we have |Gf |k+α ≤ C|f |k+(α/2) . Similarly, we can prove the desired properties for G∗ and B. Lemma 3.62
Let p > 0, q > 0 and r > 0 be such that 1 1 1 + + = 1. p q r
Then f gh ∈ L1 (Ω),
f gh1 ≤ f p gq hr
for f ∈ Lp (Ω), g ∈ Lq (Ω) and h ∈ Lr (Ω). Proof.
Let s > 0 be such that 1 1 1 = + . s p q
Using the H¨older inequality we have |f g|s dV = (|f |p )s/p (|g|q )s/q dV Ω
Ω
≤
Ω
p
|f | dV
s/p
Ω
q
|g| dV
Hence we have f g ∈ Ls (Ω),
f gs ≤ f p gq .
On the other hand, we have 1 1 + = 1. s r
s/q
.
Several Complex Variables and Integral Formulas
226
By applying the H¨ older inequality to f g ∈ Ls (Ω) and h ∈ Lr (Ω), we obtain f gh1 ≤ f gs hr ≤ f p gq hr .
Theorem 3.28 Let K(z, ζ) be a measurable function on Ω × Ω. Suppose there exist constants M > 0 and s ≥ 1 such that |K(z, ζ)|s dV (ζ) ≤ M s (z ∈ Ω). (a) Ω |K(z, ζ)|s dV (z) ≤ M s (ζ ∈ Ω). (b) Ω
Define Kf (z) =
f (ζ)K(z, ζ)dV (ζ).
Ω
Then K is a bounded operator from Lp (Ω) to Lq (Ω) with K ≤ M for all p and q satisfying 1 ≤ p, q ≤ ∞ and 1 1 1 = + − 1. q p s Proof. We prove the theorem in case 1 ≤ q < ∞, 1 < p, s < ∞. The case s = 1 will be left to the reader. Let f ∈ Lp (Ω). Since 1 p−1 s−1 + + =1 q p s and |Kf (z)| ≤
Ω
(|K|s |f |p )1/q |K|1−(s/q) |f |1−(p/q) dV (ζ),
it follows from Lemma 3.62 that 1/q s p |Kf (z)| ≤ × |K(z, ζ)| |f (ζ)| dV (ζ)| Ω
Ω
(p−1)/p (s−1)/s |K(z, ζ)|s dV (ζ) |f |p dV . Ω
Consequently we have |K(z, ζ)|s |f (ζ)|p dV (ζ) dV (z) × |Kf (z)|q dV (z) ≤ Ω
Ω
Ω
M sq(p−1)/p f pq(s−1)/s .
Integral Formulas for Strictly Pseudoconvex Domains
227
It follows from the condition (b) that pq(s−1)/s
Kf qLq ≤ M s f pLp M sq(p−1)/p f Lp
= M q f qLp .
Theorem 3.29 Let B∗ be the adjoint operator of B. Then operators B and B∗ have the following properties: (a) B is a bounded operator from Lp (Ω) to Lq (Ω) for 1 ≤ p, q ≤ ∞ and 1 1 1 q > p − 2n+2 . (b) B : L2 (Ω) → L2 (Ω) is a compact operator. (c) The kernel of B∗ is B ∗ (ζ, z) = B(z, ζ). (d) B∗ = −B. Proof. (a) follows from Theorem 3.25 and Theorem 3.28. (b) follows from Theorem 3.25 (s = 1) and Proposition A.13 in Appendix A. (c) Define ∗ f (ζ)B ∗ (z, ζ)dV (ζ). E f (z) = Ω
Using Fubini’s theorem we obtain for f, g ∈ D(Ω) (E∗ f, g) = E∗ f (z)g(z)dV (z) Ω = f (ζ)B ∗ (z, ζ)dV (ζ) g(z)dV (z) Ω Ω = f (ζ) g(z)B(ζ, z)dV (z) dV (ζ) Ω Ω = f (ζ)(Bg)(ζ)dV (ζ) Ω
= (f, Bg).
Hence E∗ = B∗ (d) Since B(ζ, z) = −B(z, ζ), we obtain f (ζ)B(z, ζ)dV (ζ) = −Bf (z). f (ζ)B ∗ (z, ζ)dV (ζ) = − B∗ f (z) = Ω
Ω
∗
Hence B = −B. Definition 3.35
For f ∈ L2 (Ω), we have a unique decomposition
f = f1 + f2
(f1 ∈ A2 (Ω), f2 ∈ (A2 (Ω))⊥ ).
Several Complex Variables and Integral Formulas
228
We define PΩ : L2 (Ω) → A2 (Ω) by PΩ f = f1 . PΩ is said to be the Bergman projection. By definition we have PΩ f = f1 ≤ f , (PΩ f, g) = (f, PΩ g). For a ∈ Ω, (3.45) shows that PΩ f (ζ)G(a, ζ)dV (ζ) PΩ f (a) = Ω
= (PΩ f, G(a, ·))
= (PΩ f, G∗ (a, ·) + B(a, ·))
= (PΩ f, G(·, a)) + (PΩ f, B(a, ·))
= (f, PΩ G(·, a)) + (PΩ f, B(a, ·)).
For fixed a, G(z, a) is holomorphic in Ω, and continuous on Ω, which implies that G(·, a) ∈ A2 (Ω). Hence we have PΩ G(·, a) = G(·, a). Therefore we obtain PΩ f (a) = (f, G∗ (a, ·)) + (PΩ f, B(a, ·)).
(3.59)
Theorem 3.30 G, G∗ are bounded operators from L2 (Ω) to L2 (Ω). Moreover, I − B : L2 (Ω) → L2 (Ω) is an invertible operator and satisfies PΩ = (I − B)−1 ◦ G∗ . Proof.
(3.60)
It follows from (3.59) that PΩ = G∗ + B ◦ PΩ . Hence we have G∗ = PΩ − B ◦ PΩ = (I − B) ◦ PΩ ,
which means that G∗ is bounded. We set T = I − B. If T (f ) = 0, then by theorem 3.29 −(f, f ) = (−Bf, f ) = (B∗ f, f ) = (f, Bf ) = (f, f ), which implies that f = 0. Therefore Ker T = {0}, and hence from Proposition A.10 in Appendix A, T : L2 (Ω) → L2 (Ω) is invertible. Hence (3.60) holds. In order to prove Theorem 3.31, we need the following two lemmas. Lemma 3.63 Let Ω ⊂⊂ Rn be an open set and let 0 < α < β < 1. Then the inclusion mapping ι : C β (Ω) → C α (Ω) is a compact operator, where C α (Ω) is the Lipschitz space of order α ( C α (Ω) is also denoted by Λα (Ω)).
Integral Formulas for Strictly Pseudoconvex Domains
229
Proof. Let {fn } be a bounded sequence in C β (Ω). Then there exists a constant M > 0 such that |fn |β,Ω ≤ M . Let ε > 0. Then any x ∈ Ω has a neighborhood Vx which satisfies the following: |fn (x) − fn (y)| 0 such that |PΩ f |k,Ω ≤ ck |f |mk ,Ω
(f ∈ C mk (Ω)).
Definition 3.37 Let Ω ⊂⊂ Cn be a domain. We say that Ω satisfies the condition (R) if Ω satisfies the condition (Rk ) for every positive integer k. The following theorem follows from Theorem 3.31. Theorem 3.32 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C ∞ boundary. Then Ω satisfies the condition (R). Theorem 3.33 Then
Let PΩ : L2 (Ω) → A2 (Ω) be the Bergman projection. (PΩ f )(z) =
f (ζ)KΩ (z, ζ)dV (ζ)
Ω
for all f ∈ L2 (Ω). Proof.
For f ∈ A2 (Ω), we have PΩ f = f . Hence we obtain
(PΩ f )(z) = (PΩ f, KΩ (·, z)) = (f, PΩ KΩ (·, z)) = (f, KΩ (·, z)). Since KΩ (z, ζ) = KΩ (ζ, z), we have the desired equality.
Integral Formulas for Strictly Pseudoconvex Domains
231
Corollary 3.8 Let Ωj , j = 1, 2, be bounded domains in Cn and let F : Ω1 → Ω2 be a biholomorphic mapping. Then PΩ1 ((f ◦ F )detF ′ ) = (detF ′ )PΩ2 (f ) ◦ F for all f ∈ L2 (Ω2 ). Proof.
For f ∈ L2 (Ω2 ), we define TF f = (f ◦ F )detF ′ . It follows that |TF f (ζ)|2 dV (ζ), |f (w)|2 dV (w) = Ω1
Ω2
2
which implies that TF f ∈ L (Ω1 ). Hence it follows from Lemma 3.64 and Theorem 3.24 that (TF f )(ζ)KΩ1 (z, ζ)dV (ζ) PΩ1 (TF f )(z) = Ω1 = (TF f )(ζ)detF ′ (z)KΩ2 (F (z), F (ζ))detF ′ (ζ)dV (ζ) Ω1 = detF ′ (z) f ◦ F (ζ)|detF ′ (ζ)|2 KΩ2 (F (z), F (ζ))dV (ζ) Ω1 ′ ˜ ˜ ˜ Ω (F (z), ζ)dV (ζ) f (ζ)K = detF (z) 2
Ω2
= detF ′ (z)PΩ2 f (F (z)).
Lemma 3.65 Let Ω be a bounded domain in Cn . For a ∈ Ω, we choose !a function ϕa ∈ D(Ω) such that ϕa depends only on |z − a| and satisfies ϕa dV = 1. Then KΩ (·, a) = PΩ ϕ¯a .
Proof. We may assume that supp(ϕa ) ⊂ B(a, ε) ⊂⊂ Ω. If f is holomorphic in Ω, then it follows from the mean value theorem that f dS (0 < ρ ≤ ε). dS = f (a) ∂B(a,ρ)
∂B(a,ρ)
Since ϕa is constant in ∂B(a, ρ), we have ϕa dS = f (a) ∂B(a,ρ)
f ϕa dS.
∂B(a,ρ)
Integrating from 0 to ε in (3.61), we obtain ϕa dV = f (a) B(a,ρ)
B(a,ρ)
f ϕa dV.
(3.61)
Several Complex Variables and Integral Formulas
232
Therefore, we have f (a) = (f, ϕa ) for any holomorphic function f in Ω. Suppose f ∈ A2 (Ω). Since PΩ f = f , we have f (a) = (f, ϕa ) = (PΩ f, ϕa ) = (f, PΩ ϕa ). Since PΩ ϕ¯a ∈ A2 (Ω), it follows from Lemma 3.52 that PΩ ϕa = KΩ (a, ·). Hence we obtain PΩ ϕa = KΩ (a, ·) = KΩ (·, a). Theorem 3.34 Ω,
Let Ω ⊂⊂ Cn satisfy the condition (Rk ). Then for a ∈ KΩ (·, a) ∈ C k (Ω).
Proof. Since the function ϕa in Lemma 3.65 belongs to C ∞ (Ω), we have ϕ¯a ∈ C mk (Ω). Hence we have KΩ (·, a) = PΩ ϕ¯a ∈ C k (Ω). Lemma 3.66 Let Ω ⊂⊂ Rn be a domain with C k boundary, k ≥ 1. For a ∈ Ω, we denote by ϕ!a any function which depends only on |z − a| and satisfies ϕa ∈ Cc∞ (Ω), ϕa dV = 1. We set R(Ω) = {α | α is a finite linear combination of ϕa }.
If f ∈ C k (Ω) satisfies the conditions (∂ α f )(x) = 0
(x ∈ ∂Ω, |α| ≤ k),
then f is a limit of functions in R(Ω) in the C k (Ω) norm. Proof. First we show that f is a limit of functions in D(Ω) in the C k (Ω) norm. Using a partition of unity argument, there exists a neighborhood U of P ∈ ∂Ω such that if we denote n the unit inward normal vector at P , then supp(f ) ⊂ U and f (x − τ n) has a compact support in Ω ∩ U for any sufficiently small τ > 0. We define f˜ such that f˜(x) = f (x) for x ∈ Ω ∩ U , f˜(x) = 0 for x ∈ U − Ω. Then by the assumption, f˜ ∈ Cck (U ). If we set f˜τ (x) = f˜(x − τ n), then |f − f˜τ |k,Ω∩U → 0 as τ → 0. Thus we may assume k !that f ∈ Cc (Ω). Suppose ϕ ∈nD(B(0, 1)) depends only on |z| and satisfies ϕdV = 1. We set ϕj (x) = j ϕ(jx), j = 1, 2, · · · , and fj (x) = f (y)ϕj (x − y)dV (y) = f (x − y)ϕj (y)dV (y). Rn
Rn
Integral Formulas for Strictly Pseudoconvex Domains
233
Then fj ∈ D(Ω) for any sufficiently large j . Moreover, we have ∂ α fj (x) = f (y)∂xα ϕj (x − y)dV (y) = (∂ α f )(y)ϕj (x − y)dV (y), Rn
Rn
which implies that |f − fj |k,Ω → 0
(j → ∞).
Each ∂ α fj (x) is a limit of Riemann sums N
ν=1
cν f (ην )∂xα ϕj (x − ην ),
(3.62)
where cν are positive constants. There exists a constant M > 0 such that |cν f (ην )∂xα ϕj (x − ην )| ≤ M cν |f (ην )|. ! Since N |f |dV , and hence converge. ν=1 cν |f (ην )| are Riemann sums of Hence (3.62) converge to fj uniformly on Ω. Therefore, if we set gN (x) =
N
ν=1
cν f (ην )ϕj (x − ην ),
then |fj − gN |k,Ω → 0 as N → ∞, and hence gN ∈ R(Ω).
Now we prove the following lemma (see Bell-Ligocka [BEL]). Lemma 3.67 (Bell’s density lemma) Let Ω be a bounded domain in C n with C ∞ boundary. Let Ω satisfy the condition (R0 ). Then given u ∈ C k+1 (Ω), there exists a function g ∈ C k (Ω) with PΩ g = 0 and such that ¯
∂ αβ (u − g)|∂Ω = 0
(|α| + |β| ≤ k).
Proof. Let ρ be a defining function for Ω. There exist a constant C > 0, Pj ∈ ∂Ω, j = 1, · · · , N , and neighborhoods Uj of Pj with the following properties: N
(1) ∂Ω ⊂ ∪ Ui i=1
(2) For each i (1 ≤ i ≤ N ), there exists integer j with 1 ≤ j ≤ n such that ∂ρ (z ∈ Ui ). ∂zj (z) > C
234
Several Complex Variables and Integral Formulas
N Let {αj }N j=1 be a partition of unity subordinate to {Uj }j=1 , that is, N {αj }j=1 satisfies the following properties:
(a) αj ∈ C ∞ (Cn ). (b) supp(αj ) ⊂⊂ Uj . N (c) There exists a neighborhood U of ∂Ω such that j=1 αj (z) = 1 for z ∈ U. It is sufficient to show that Lemma 3.67 holds for αi u instead of u. So we rewrite αi u by u. Thus we have supp(u) ⊂ {w ∈ Ω |
∂ρ (w) = 0}. ∂z1
Define w1 (z) =
u(z)ρ(z) , ∂ρ ∂z1 (z)
v1 (z) =
∂w1 (z). ∂z1
Then we have ∂ v1 (z) = u(z) + ρ(z) ∂z
u(z) ∂ρ ∂z1 (z)
,
and hence v1 − u = 0 on ∂Ω. Using the fact that w1 |∂Ω = 0, KΩ (·, z) is holomorphic in Ω and KΩ (·, z) ∈ C(Ω), we obtain KΩ (ζ, z) w1 (ζ) dV (ζ) = 0. PΩ v1 (z) = v1 (ζ)KΩ (ζ, z)dV (ζ) = − ∂ ζ¯j Ω Ω Hence Lemma 3.67 holds in case k = 0. We assume that wk−1 and vk−1 = ∂wk−1 ∂z1 have already been constructed and that vk−1 is equal to u on ∂Ω up to derivatives of order k − 2 and satisfies PΩ vk−1 = 0. Define a differential operator D on Ω by 2n ∂ρ ∂ϕ ν=1 ∂xν (z) ∂xν (z) . D(ϕ)(z) = |∇ρ(z)|2 Define wk and vk by wk = wk−1 + θk ρk ,
vk =
∂wk , ∂z1
where θk is defined by θk =
Dk−1 (u − vk−1 ) ∂ρ k! ∂z
.
Integral Formulas for Strictly Pseudoconvex Domains
235
Since vk = vk−1 +
∂ (θk ρk ), ∂z
we obtain Dk−1 (u − vk ) = Dk−1 (u − vk−1 ) − Dk−1
∂ (θk ρk ) . ∂z
(3.63)
Further, we devide the second term of the right side in (3.63) into a term which involves ρ and a term which does not involve ρ. Then we have a representation ∂ρ ∂ k! + (the term involving ρ). (θk ρk ) = θk Dk−1 ∂z ∂z1 Consequently, Dk−1 (u − vk ) = Dk−1 (u − vk−1 ) − θk
∂ρ k! + (the term involving ρ). ∂z
By the definition of θk , we have Dk−1 (u − vk )|∂Ω = 0. Next we choose vector fields τ1 , · · · , τ2n−1 at P ∈ ∂Ω such that {τ1 , · · · , τ2n−1 , D} are orthogonal basis at P . Every vector field at P ∈ ∂Ω is denoted by a linear combination of D and τi , i = 1, · · · , 2n − 1. For simplicity, we denote all τi , i = 1, · · · , 2n − 1, by τ . Then, in order that vk and u coincide on ∂Ω up to derivatives of order k − 1, it is sufficient to show that if s + t ≤ k − 1, then τ s Dt (u − vk )|∂Ω = 0. In case s + t < k − 1, by the inductive hypothesis we have ∂ (θk ρk ) = 0. τ s Dt (u − vk )|∂Ω = τ s Dt (u − vk−1 ) + Dt ∂z ∂Ω
In case s = 0, t = k − 1, we have already proved. In case s + t = k − 1, s ≥ 1, we have τ s Dt (u − vk )|∂Ω = 0.
By Lemma 1.21, there exists a C ∞ function h in a neighborhood of P ∈ ∂Ω such that τ s−1 Dt (u − vk ) = ρh.
Several Complex Variables and Integral Formulas
236
If τ has a representation in a neighborhood U of P τ=
2n X
aj (z)
j=1
∂ , ∂xj
then we obtain τ (ρ(z)h(z))|∂Ω∩U =
2n X
aj (z)
j=1
∂ρ (z)h(z)|∂Ω∩U = 0, ∂xj
which means that vk and u coincide on ∂Ω up to derivatives of order k − 1. By the definition of wk , we have wk = γρ (γ is of class C ∞ on Ω), and hence PΩ vk = 0. Lemma 3.67 is proved. Theorem 3.35 Let Ω be a bounded domain in C n with C ∞ boundary and satisfy the condition (R1 ). Suppose f is a holomorphic function in Ω with f ∈ C ∞ (Ω). Then f is a limit of a sequence of finite linear combinations of the elements in {KΩ (·, a) | a ∈ Ω} in C 1 (Ω) norm. Proof. Let f be holomorphic in Ω and f ∈ C ∞ (Ω). Then we have f = PΩ f . By Bell’s density theorem (Lemma 3.67), there exists g ∈ C m1 (Ω) such that f − g is equal to 0 on ∂Ω up to derivatives of order m1 . Moreover we have f = PΩ (f − g). By Lemma 3.66 there exist gk ∈ Cc (Ω), k = 1, 2, · · · , such that gk (z) =
Nk X
(k)
(k)
αj ϕak ,
j=1
j
(k)
and f − g is the limit of {gk } in C m1 (Ω) norm, where ϕak ∈ Cc∞ (Ω) depend j R (k) only on |z − aakj | and satisfy ϕak dV = 1. By Lemma 3.65, we obtain j
PΩ gk =
Nk X
(k)
(k)
αj PΩ ϕak (z) =
j=1
j
Nk X
(k)
αj KΩ (z, akj ).
j=1
By the condition (R1 ), we have |PΩ (f − g − gk )|1,Ω ≤ c1 |f − g − gk |m1 ,Ω , which implies that |f −
Nk X j=1
(k)
αj KΩ (z, akj )|1,Ω ≤ c1 |f − g − gk |m1 ,Ω .
Integral Formulas for Strictly Pseudoconvex Domains
Definition 3.38 condition (Bk ) if
237
Let Ω be a domain in Cn . We say that Ω satisfies the
(a) For any a ∈ Ω, KΩ (·, a) ∈ C k (Ω). (b) For any P ∈ Ω, there exist a0 , a1 , · · · , an ∈ Ω such that KΩ (P, a0 ) = 0,
(3.64)
and K (P, a ) · · · K (P, a ) 0 Ω n Ω ∂KΩ (P, a ) · · · ∂KΩ (P, a ) ∂z1 0 n ∂z1 = 0, .. .. .. . . . ∂KΩ ∂K ∂z (P, a0 ) · · · ∂z Ω (P, an ) n n
(3.65)
where the derivatives in (3.65) are taken with respect to the first variable in KΩ (·, aj ). Theorem 3.36 Let Ω ⊂⊂ Cn be a domain with C ∞ boundary. If Ω satisfies the condition (Rk ), then Ω satisfies the condition (Bk ). Proof. The condition (Bk ) (a) follows from Theorem 3.34. We will show (Bk ) (b). We fix P ∈ Ω. Suppose (3.65) is equal to 0 for all (a0 , a1 , · · · , an ) ∈ Ωn+1 . For any g0 , g1 , · · · , gn ∈ O(Ω)∩C ∞ (Ω) and any ε > 0, by Theorem 3.35, there exist akj ∈ Ω, j = 1, · · · , Nk , and constants bkj , j = 1, · · · , Nk , such that |gk −
Nk
bkj KΩ (·, akj )|1,Ω < ε.
j=1
We set αk (z) =
Nk
bkj KΩ (z, akj ).
j=1
Then by the assumption we have α ··· α n 0 ∂α0 · · · ∂αn ∂z1 ∂z1 . .. .. = 0. . . . . ∂α0 n ∂z · · · ∂α ∂zn n
238
Several Complex Variables and Integral Formulas
On the other hand we g0 ∂g0 ∂z1 . . . ∂g0 ∂z n
have · · · gn α0 · · · αn ∂α0 · · · ∂αn n · · · ∂g ∂z1 ∂z1 ∂z1 − .. .. .. .. .. = O(ε). . . . . . ∂α0 ∂αn n · · · · · · ∂g ∂zn ∂zn ∂zn
Since ε > 0 is arbitrary, we have for any g0 , g1 , · · · , gn ∈ O(Ω) ∩ C ∞ (Ω) g0 · · · gn ∂g0 · · · ∂gn ∂z1 . . ∂z. 1 = 0. . . . . . . ∂g0 n ∂z · · · ∂g ∂zn n
If we set g0 = 1, gk (z) = zk for k = 1, · · · , n, then the left side of the above equality is 1, which is a contradiction. This proves (Bk ) (b), which completes the proof of Theorem 3.36. Corollary 3.9 Let Ω ⊂⊂ Cn be a domain with C ∞ boundary and satisfy the condition (Bk ). Then for any P ∈ ∂Ω there exist a neighborhood W of P and a0 , a1 , · · · , an ∈ Ω such that if we set uj (z) =
KΩ (z, aj ) , KΩ (z, a0 )
u = (u1 , · · · , un ),
then uj ∈ O(W ∩ Ω) ∩ C k (W ∩ Ω), and det u′ (P ) = 0.
(3.66)
Moreover, each component of the inverse mapping u−1 : u(W ∩Ω) → W ∩Ω belongs to O(u(W ∩ Ω)) ∩ C k (u(W ∩ Ω)). Proof. We fix P ∈ ∂Ω. Since Ω satisfies the condition (Bk ), by Theorem 3.36 there exist a0 , a1 , · · · , an ∈ Ω which satisfy (3.64) and (3.65). Hence there exists a neighborhood W of P such that KΩ (z, a0 ) = 0 for z ∈ W ∩ Ω, which implies that uj ∈ O(W ∩ Ω) ∩ C k (W ∩ Ω). Since ∂uj ∂KΩ ∂KΩ (P ) = KΩ (P, a0 )−1 (P, aj ) − (P, a0 )KΩ (P, a0 )−2 KΩ (P, aj ), ∂zk ∂zk ∂zk
Integral Formulas for Strictly Pseudoconvex Domains
239
we have K (P, a ) · · · K (P, a ) 0 Ω n Ω ∂KΩ (P, a ) · · · ∂KΩ (P, a ) ∂z1 0 n ∂z1 .. .. .. . . . ∂KΩ ∂K ∂z (P, a0 ) · · · ∂z Ω (P, an ) n n KΩ (P, a0 ) KΩ (P, a0 ) KΩ (P, a0 ) · · · KΩ (P, an ) ∂un ∂u1 ··· 0 ∂z ∂z 1 1 n = K(P, a0 ) .. .. .. .. . . . . ∂u ∂u n 1 · · · 0 ∂zn ∂zn ∂u1 n ∂z · · · ∂u ∂z1 1 .. . = K(P, a0 )n+1 ... ... . ∂u1 · · · ∂un ∂zn ∂zn This proves (3.66). Since uj is holomorphic in W ∩ Ω, we have ∂uj (P ) = 0. ∂ z¯j Thus the Jacobian of u at P Ju(P ) = |det u′ (P )|2 = 0. Since uj can be extended to C k functions in W , by contracting W if necessary, u : W → u(W ) is a C k diffeomorphism. Since u : W ∩Ω → u(W ∩Ω) is a holomorphic mapping, u−1 : u(W ∩ Ω) → W ∩ Ω is a holomorphic mapping. Theorem 3.37 Let Ω1 and Ω2 be bounded domains in Cn and satisfy the condition (Bk ) for k ≥ 1. Then every holomorphic mapping F : Ω1 → Ω2 belongs to C k (Ω1 ). Proof. The proof involves three steps. [1] There exist constants c1 , c2 such that 0 < c1 ≤ |det F ′ (z)| ≤ c2 for all z ∈ Ω1 . (Proof of [1]) We assume that there does not exist c1 which satisfies 0 < c1 ≤ |det F ′ (z)| for all z ∈ Ω1 . Then there exists a sequence {pν } ⊂ Ω1 such that det F ′ (pν ) → 0 as ν → 0. Taking the subsequence of {pν }, we may assume that {pν } converges. We set limν→∞ pν = P . Then P ∈ ∂Ω1 . By Theorem 3.24, for a ∈ Ω1 we have KΩ1 (pν , a) = det F ′ (pν )KΩ2 (F (pν ), F (a))det F ′ (a).
(3.67)
From the condition (Bk ), KΩ1 (·, a) and KΩ2 (·, F (a)) are continuous on Ω1 and Ω2 , respectively, and hence KΩ1 (P, a) = 0 for a ∈ Ω1 . This contradicts
Several Complex Variables and Integral Formulas
240
the condition (Bk ). By adopting the same argument to F −1 , there exists c > 0 such that |det (F −1 )′ (w)| ≥ c for w ∈ Ω2 . Hence we have |det F ′ (z)| ≤ 1/c for z ∈ Ω1 . [2] F ∈ C(Ω1 ). ∂f (Proof of [2]) It is sufficient to show that ∂zji , i, j = 1, · · · , n, are bounded in Ω1 . Suppose there exists {pν } ⊂ Ω1 such that ∂fj max (pν ) → ∞ (ν → ∞). j,i ∂zi
(3.68)
We set qν = F (pν ). Taking subsequences, we may assume that {pν } and {qν } converge. We set limν→∞ pν = P (P ∈ ∂Ω1 ), limν→∞ qν = Q. Then Q ∈ ∂Ω2 . For Q ∈ ∂Ω2 , we choose b0 , b1 , · · · , bn ∈ Ω2 satisfying the condition (Bk ). Then we have KΩ2 (Q, b0 ) = 0. We define v = (v1 , · · · , vn ) by vj (w) =
KΩ2 (w, bj ) KΩ2 (w, b0 )
(j = 1, · · · , n).
By Corollary 3.9 we have det v ′ (Q) = 0 and there exists a neighborhood W1 of Q such that each component of v −1 belongs to C k (W1 ∩ Ω1 ). If we set aj = F −1 (bj ) for j = 0, 1, · · · , n, then KΩ1 (P, a0 ) = 0 by substituting a = a0 into (3.67). Thus there exists a neighborhood W2 of P such that if we set uj (z) =
KΩ1 (z, aj ) , KΩ1 (z, a0 )
then we have uj ∈ O(W1 ∩ Ω1 ) ∩ C k (W1 ∩ Ω1 ). For a sufficiently large ν0 , we have pν ∈ W1 ∩ Ω1 , qν ∈ W2 ∩ Ω2 . By Theorem 3.24 and the definition of uj and vj , we obtain uj (z) = vj (F (z))λj
(z ∈ W1 ∩ Ω1 , F (z) ∈ W2 ∩ Ω2 ),
where we define λj =
det F ′ (aj ) . det F ′ (a0 )
(3.69)
Integral Formulas for Strictly Pseudoconvex Domains
241
We set
λ1 0 · · · 0 .. 0 λ2 . . Λ= .. .. . . 0 0 · · · λn
It follows from (3.69) that
u(z) = Λv(F (z)).
(3.70)
By the chain rule we have u′ (pν ) = Λv ′ (F (pν ))F ′ (pν ), which implies that F ′ (pν ) = v ′ (qν )−1 Λ−1 u′ (pν ). Since v ′ is a coordinate system in a neighborhood of Q, each component of v ′ (qν )−1 is bounded. Further, each component of u′ is bounded in a neighborhood of P . Hence each component of F ′ (pν ) remains bounded as ν → ∞, which contradicts (3.68). [3] F ∈ C k (Ω1 ). (Proof of [3]) It follows from (3.70) that for z ∈ W1 ∩ Ω1 , we have F (z) = v −1 (Λ−1 (u(z))), which means that F ∈ C k (W1 ∩ Ω1 ). Corollary 3.10 (Fefferman’s mapping theorem) Let Ω1 and Ω2 are strictly pseudoconvex domains in Cn with C ∞ boundary. Then every biholomorphic mapping F : Ω1 → Ω2 belongs to C ∞ (Ω1 ). Proof. Since Ω1 and Ω2 satisfy the condition (Rk ) for any positive integer k by Theorem 3.33, Ω satisfies the condition (Bk ). Hence it follows from Theorem 3.37 that F ∈ C k (Ω1 ).
Exercises 3.1
Prove the following: Let δ > 0 and Γδ = {(x1 , x′ ) ∈ RN | 0 < x1 < δ, |x′ | < δ}, and suppose g ∈ C 1 (Γδ ) satisfies |dg(x)| ≤ Kxα−1 1
242
Several Complex Variables and Integral Formulas
for x ∈ Γδ . Then there is a constant C depending only on α and δ such that |g(x) − g(y)| ≤ CK|x − y|α for x, y ∈ Γδ/2 with |x − y| ≤ δ/2. 3.2
3.3
3.4
Let Ω be a convex domain in Cn and let F1 : Ω → C be a C 1 function in Ω. For w, z ∈ Ω and 1 < θ, λ ≤ 1, dF1 (z + λθ(w − z)) dF1 (z + θ(w − z)) . λ=1 = θ dλ dθ
Let H be a Hilbert space and let ϕ be a continuous linear functional on H. Define M = {x ∈ H | ϕ(x) = 0}. Show that if M = H, then M ⊥ is one dimensional. Let Ω = {z ∈ C | |z| < 1}. Define
√ π √ ϕn (z) = z n. n+1
Prove that {ϕn } is a complete orthonormal sequence in A2 (Ω). 3.5
Let Ω = {z ∈ C | |z| < 1}. Prove that the Bergman kernel KΩ (z, ζ) for Ω is given by KΩ (z, ζ) =
3.6
1 1 ¯ 2. π (1 − z ζ)
Let Ω ⊂⊂ Cn be a domain. We set kΩ (z) = KΩ (z, z). Show that
∂KΩ (z, ζ) ∂f = (a) f (z), (ζ) ( f ∈ A2 (Ω), ζ ∈ Ω). ∂ζν ∂ ζ¯ν (z ∈ Ω). (b) kΩ (z) > 0 (c) log kΩ (z) is strictly plurisubharmonic in Ω. 3.7 (Bergman metric) Let Ω ⊂⊂ Cn . The Hermitian metric for Ω is defined by gij (z) =
∂ 2 log kΩ (z). ∂zj ∂ z¯k
Integral Formulas for Strictly Pseudoconvex Domains
243
Let γ : [0, 1] → Ω be a C 1 curve. The length of γ with respect to the Bergman metric |γ|B(Ω) is defined by 1/2 1 n |γ|B(Ω) = gij (γ(t))γi′ (t)γj (t) dt. 0
i,j=1
For z1 , z2 ∈ Ω, we define the distance of z1 , z2 with respect to the Bergman metric by δB(Ω) (z1 , z2 ) = inf |γ|B(Ω) , γ
where the infimum is taken for all C 1 curves in Ω which connect z1 and z2 . Prove that if f : Ω1 → Ω2 is a biholomorphic mapping, then δB(Ω1 ) (z1 , z2 ) = δB(Ω2 ) (f (z1 ), f (z2 )).
This page intentionally left blank
Chapter 4
Integral Formulas with Weight Factors
In this chapter we study the Berndtsson-Andersson formula on bounded domains in Cn with smooth boundary and the Berndtsson formula on submanifolds in general position of bounded domains in Cn with smooth boundary. By applying the Berndtsson-Andersson formula, we prove Lp estimates for the ∂¯ problem in strictly pseudoconvex domains in Cn with smooth boundary. Moreover, using the Berndtsson formula we give two counterexamples of Lp (2 < p ≤ ∞) extensions of bounded holomorphic functions from submanifolds of complex ellipsoids due to Mazzilli [MAZ1] and Diederich-Mazzilli [DIM1]. Finally, we give the alternative proof of the bounded extension of holomorphic functions from affine linear submanifolds of strictly convex domains using the method of Diederich-Mazzilli [DIM2].
4.1
The Berndtsson-Andersson Formula
In this section we study the integral formula obtained by BerndtssonAndersson [BRA]. Let µ =< ξ, η >−n ω ′ (ξ) ∧ ω(η) be a differential form in Cn × Cn = {(ξ, η) | ξ ∈ Cn , η ∈ Cn }, where we define ′
ω (ξ) :=
n j=1
(−1)j−1 ξj ∧ dξi , i=j
245
ω(η) := dη1 ∧ · · · ∧ dηn ,
Several Complex Variables and Integral Formulas
246
< ξ, η >:=
n
ξj ηj .
j=1
By Lemma 3.2, if < ξ, µ >= 0, then dµ = 0. Let Ω be a bounded domain in Cn with smooth boundary. Assume that a C 1 mapping s = (s1 , · · · , sn ) : Ω × Ω → Cn satisfies the following conditions: (A) If ζ = z, then < s(z, ζ), ζ − z >= 0. (B) For any compact set K ⊂ Ω, there exist constants C1 = C1 (K) > 0, C2 = C2 (K) > 0 such that |s(z, ζ)| ≤ C1 |ζ − z|,
| < s(z, ζ), ζ − z > | ≥ C2 |ζ − z|2
for ζ ∈ Ω and z ∈ K. In what follows we assume that s satisfies the above conditions (A) and (B). Define ψ : Ω × Ω\∆ → E to be ψ(z, ζ) = (s(z, ζ), ζ − z), where ∆ = {(z, z) | z ∈ Cn }. Let K be the pullback of µ by ψ. Then 1 ω ′ (s) ∧ ω(ζ − z) < s, ζ − z >n n 1 = (−1)j−1 sj ∧ dz,ζ si i=j < s, ζ − z >n j=1
K = ψ∗ µ =
∧(dζ1 − dz1 ) ∧ · · · ∧ (dζn − dzn ).
We denote by Kp,q the component of K which is of degree (p, q) with respect to z and of degree (n − p, n − q − 1) with respect to ζ. Then we have the following theorem. 1 (Ω), one has Theorem 4.1 For f ∈ C(p,q) (a) For q > 0, p+q+1 ¯ ¯ , ∂f ∧ Kp,q − ∂z f =C f ∧ Kp,q + (−1) f ∧ Kp,q−1 Ω
∂Ω
Ω
where C = Cp,q,n is a numerical constant depending only on p, q, n. (b) For q = 0, p+1 ¯ ∂f ∧ Kp,0 , f ∧ Kp,0 + (−1) f =C ∂Ω
Ω
where C = Cp,n is a numerical constant depending only on p, n.
Integral Formulas with Weight Factors
247
Proof. Let ϕ be a C ∞ (n − p, n − q) form in Ω with compact support. For ε > 0, we set Uε = {(ζ, z) ∈ Ω × Ω | |ζ − z| < ε} Ωε = Ω × Ω − U ε . If we choose ε sufficiently small in comparison with the distance between supp(ϕ) and ∂Ω, then ∂Ωε ∩ (Cn × supp(ϕ)) = {(∂Ω × Ω) ∪ ∂Uε } ∩ (Cn × supp(ϕ)). It follows from Stokes’ theorem that ϕ∧f ∧K dz,ζ (ϕ(z) ∧ f (ζ) ∧ K(z, ζ)) = ∂Ωε Ωε = ϕ∧f ∧K− ∂Ω×Ω
Since dK = ψ ∗ dµ = 0, we obtain dϕ ∧ f ∧ K + (−1)p+q ϕ ∧ df ∧ K = Ωε
∂Ω×Ω
Ωε
∂Uε
ϕ∧f ∧K −
ϕ ∧ f ∧ K.
∂Uε
ϕ ∧ f ∧ K. (4.1)
Since K=O
|s| | < s, ζ − z > |n
= O(|ζ − z|1−2n ),
the two integrals in the left side of (4.1) converges as ε → 0. Next we investigate the second integral in the right side of (4.1). We obtain ω ′ (s) ∧ ω(ζ − z) lim . (4.2) ϕ ∧ f ∧ K = lim ϕ∧f ∧ ε→0 ∂U ε→0 ∂U < s, ζ − z >n ε ε Lemma 3.5 implies that ω ′ (s) ∧ ω(ζ − z) < s, ζ − z >−n is invariant when
we replace s by s || . Hence we may assume that < s, ζ − z >> 0 for ζ = z. Define b = ζ¯ − z¯,
sλ = λs + (1 − λ)b
(0 < λ < 1).
Further we define h : Ω × Ω × [0, 1] → Cn × Cn by h(z, ζ, λ) = (sλ (z, ζ), ζ − z).
Several Complex Variables and Integral Formulas
248
If we set H = h∗ µ, then we obtain H(z, ζ, λ) =< λs + (1 − λ)b, ζ − z >−n ω ′ (λs + (1 − λ)b) ∧ ω(ζ − z). We set Iε =
∂({|ζ−z|=ε}×[0,1])
ϕ(z) ∧ f (ζ) ∧ H(z, ζ, λ).
Since dH = h∗ dµ = 0, it follows from Stokes’ theorem that Iε = d(ϕ ∧ f ) ∧ H. {|ζ−z|=ε}×[0,1]
Since supp(ϕ) is compact, ϕ = 0 on ∂{|ζ − z| = ε} for any sufficiently small ε > 0. Hence we obtain ϕ ∧ f ∧ H(z, ζ, 0). (4.3) ϕ ∧ f ∧ H(z, ζ, 1) − Iε = |ζ−z|=ε
|ζ−z|=ε
We denote by H ′ the component of H which involves dλ. Then |sλ |(|s| + |b|) |H ′ | ≤ C = O(|ζ − z|2−2n ). λ < s, ζ − z > +(1 − λ)|ζ − z|2 )n Consequently we have limε→0 Iε = 0. It follows from (4.2) and (4.3) that ϕ(z) ∧ f (ζ) ∧ H(z, ζ, 0). (4.4) ϕ ∧ f ∧ K = lim lim ε→0
ε→0
∂Uε
∂Uε
By the same method as the proof of Theorem 3.3, the right side of (4.4) is ! equal to Cp,q,n Ω ϕ ∧ f . Letting ε → 0 in (4.1) we obtain p+q dϕ∧f ∧K +(−1) ϕ∧df ∧K = ϕ∧f ∧K −Cp,q,n ϕ∧f Ω×Ω
Ω×Ω
∂Ω×Ω
Ω
(4.5)
and
dϕ(z) ∧ f (ζ) ∧ K(z, ζ) = f (ζ) ∧ K(z, ζ) dϕ(z) ∧ ζ∈Ω z∈Ω p+q+1 = (−1) ϕ(z) ∧ dz f (ζ) ∧ K(z, ζ). Ω×Ω
z∈Ω
ζ∈Ω
Integral Formulas with Weight Factors
249
Since ϕ(z) ∧ f (ζ) is of degree n with respect to dζ and dz and that K(z, ζ) is of degree ≥ n with respect to dζ and dz, we have dϕ(z)∧f (ζ)∧K(z, ζ) = (−1)p+q+1 ϕ(z)∧ ∂¯z f (ζ)∧K(z, ζ). Ω×Ω
z∈Ω
ζ∈Ω
Thus by (4.5), we obtain p+q ¯ ¯ ϕ∧ f ∧ K = (−1) ϕ∧ ∂f ∧ K − ∂z f ∧K Ω ∂Ω Ω Ω Ω +Cp,q,n ϕ ∧ f. Ω
! This proves (a). In case q = 0, ∂¯z Ω f ∧ K is of degree ≥ 1 with respect to d¯ z , which implies that ¯ ∧ K + Cp,n ∂f ϕ ∧ f. f ∧ K = (−1)p ϕ∧ ϕ∧ Ω
∂Ω
Ω
Ω
This proves (b).
Ω
Corollary 4.1 Assume that in addition to the conditions (A) and (B), s : Ω × Ω → Cn satisfies the conditon: (C) For ζ ∈ ∂Ω, s(z, ζ) is holomorphic with respect to z ∈ Ω. Then p+q u(z) = (−1) Cp,q,n f (ζ) ∧ Kp,q−1 (z, ζ) Ω
¯ = 0. ¯ = f for f ∈ C 1 (Ω) (q > 0) with ∂f is a solution of the equation ∂u (p,q)
Proof. It follows from the condition (C) that K(z, ζ) is of degree 0 with respect to d¯ z for ζ ∈ ∂Ω, and hence Kp,q = 0. Therefore, Corollary 4.1 follows from Theorem 4.1 (a). Next we study the differential form A = exp < ξ, η > ω(ξ) ∧ ω(η) in Cn × Cn . Suppose a C 1 mapping Q(z, ζ) : Ω × Ω → Cn is holomorphic in z ∈ Ω for ζ fixed. Let Q = (Q1 , · · · , Qn ). We define ψ : (Ω × Ω\∆) × (0, ∞) → Cn × Cn by ψ(z, ζ, t) = (Q(z, ζ) + ts(z, ζ), ζ − z). We set N = ψ ∗ A. Then N can be written N = exp < Q + ts, ζ − z > d(Q1 + ts1 ) ∧ · · · ∧ d(Qn + tsn ) ∧ ω(ζ − z). (4.6)
Several Complex Variables and Integral Formulas
250
We write N = Nt + N ′ , where Nt is the component of N which contains dt, and N ′ is the the component of N which does not contain dt. Then Nt = −exp < Q, ζ − z > exp{t < s, ζ − z >}× {tn−1 ω ′ (s) ∧ ω(ζ − z) ∧ dt +
n−2 k=0
tk ak ∧ dt},
where ak are differential forms which do not contain t. Since dA = 0, we have dN = ψ ∗ dA = 0. Consequently, 0 = dζ,z,t N = dζ,z Nt + dt N ′ + dζ,z N ′ .
(4.7)
Since the last term in the right side of (4.7) does not contain dt, we obtain dζ,z Nt = −dt N ′ .
(4.8)
In the moment we assume that Re < s, ζ − z >< 0 for ζ = z. Later we show that this assumption is not necessary. It follows from (4.8) that ∞ ∞ dζ,z Nt = − dt N ′ = N ′ |t=0 dζ,z K = 0
0
= exp < Q, ζ − z > ω(Q) ∧ ω(ζ − z).
We set P = exp < Q, ζ − z > ω(Q) ∧ ω(ζ − z). Then we have dζ,z K = P.
(4.9)
Since Q is holomorphic in z, dQ does not contain d¯ zj , and hence P does not contain d¯ zj . By the integration by parts, we obtain
0
∞
∞ et n−1 t < s, ζ − z > 0 ∞ t e − (n − 1)tn−2 dt < s, ζ − z > 0 1 = · · · = (−1)n (n − 1)! . < s, ζ − z >n
et tn−1 dt =
Integral Formulas with Weight Factors
251
Hence K is expressed by K = (−1)n−1 (n − 1)!exp < Q, ζ − z >
+
n−2 k=0
ω ′ (s) ∧ ω(ζ − z) < s, ζ − z >n
(4.10)
O(< s, ζ − z >−(k+1) ).
1 Theorem 4.2 For f ∈ C(p,q) (Ω), one has (a) In case q > 0, p+q+1 ¯ ¯ , ∂f ∧ Kp,q − ∂z f ∧ Kp,q−1 f ∧ Kp,q + (−1) f =C ∂Ω
Ω
Ω
where Kp,q are components of K which are (p, q) forms with respect to z and (n − p, n − q − 1) forms with respect to ζ. (b) In case q = 0, p+1 ¯ ∂f ∧ Kp,0 − f ∧ Kp,0 + (−1) f ∧ Pp,0 . f =C ∂Ω
Ω
Ω
Proof. Let Ωε , Uε and ϕ be the same notations as in the proof of Theorem 4.1. It follows from Stokes’ theorem that ϕ∧f ∧K dz,ζ (ϕ(z) ∧ f (ζ) ∧ K(z, ζ)) = Ωε ∂Ωε = ϕ ∧ f ∧ K. ϕ∧f ∧K− ∂Uε
∂Ω×Ω
By (4.9) we obtain dϕ ∧ f ∧ K + (−1)p+q Ωε
Ωε
=
∂Ω×Ω
ϕ ∧ df ∧ K +
ϕ∧f ∧K −
∂Uε
Ωε
ϕ∧f ∧P
(4.11)
ϕ ∧ f ∧ K.
On the other hand, we have |s| = O(|ζ − z|1−2n ), K=O | < s, ζ − z > |n which means that the three integrals in the left side of (4.11) converge as ε → 0. Next we investigate the second integral in the right side of (4.11).
Several Complex Variables and Integral Formulas
252
It follows from (4.10) that K = −(n − 1)!(exp < Q, ζ − z >)
ω ′ (s) ∧ ω(ζ − z) + T1 , < s, z − ζ >n
where T1 = O(|ζ − z|2−2n ). Since exp < Q, ζ − z >= 1+ < Q, ζ − z > +
< Q, ζ − z >2 + ··· , 2!
K can be written K = −(n − 1)!
ω ′ (s) ∧ ω(ζ − z) + T2 , < s, z − ζ >n
where T2 = O(|ζ − z|2−2n ). Consequently, using the same method as in the proof of Theorem 4.1 we obtain ϕ∧f ∧K lim ε→0 ∂U ε ω ′ (s) ∧ ω(ζ − z) = (−1)n−1 (n − 1)! lim ϕ∧f ∧ ε→0 ∂U < s, ζ − z >n ε = Cp,q,n ϕ ∧ f, Ω
where Cp,q,n are numerical constants depending only on p, q, n. Letting ε → 0 in (4.11) we obtain dϕ ∧ f ∧ K + (−1)p+q ϕ ∧ f ∧ P (4.12) ϕ ∧ df ∧ K + Ω×Ω
Ω×Ω
Ω×Ω
= Consequently,
∂Ω×Ω
ϕ ∧ f ∧ K − Cp,q,n
Ω
ϕ ∧ f.
dϕ(z) ∧ f (ζ) ∧ K(z, ζ) = f (ζ) ∧ K(z, ζ) dϕ(z) ∧ ζ∈Ω z∈Ω p+q+1 = (−1) ϕ(z) ∧ dz f (ζ) ∧ K(z, ζ). Ω×Ω
z∈Ω
ζ∈Ω
Integral Formulas with Weight Factors
253
Using the fact that ϕ(z) ∧ f (ζ) is of degree n with respect to dζ and dz and that K(z, ζ) is of degree ≥ n with respect to dζ and dz, we have Z Z Z dϕ(z)∧f (ζ)∧K(z, ζ) = (−1)p+q+1 ϕ(z)∧ ∂¯z f (ζ)∧K(z, ζ). Ω×Ω
z∈Ω
ζ∈Ω
It follows from (4.12) that Z Z Z Z Z ¯ ∧ K − ∂¯z ϕ∧ f ∧ K = (−1)p+q ϕ∧ ∂f f ∧K Ω ∂Ω Ω Ω Z ZΩ Z + ϕ∧ f ∧ P + Cp,q,n ϕ ∧ f. Ω
Ω
Ω
Since Q(z, ζ) is holomorphic in z ∈ Ω, ω(Q) is of degree 0 with respect to d¯ z , which implies that Pp,q = 0 if q > 0. This proves (a). If q = 0, then ϕ(z) is of degree n with respect to d¯ z , which means that Z ϕ(z) ∧ ∂¯z f (ζ) ∧ K(z, ζ) = 0. Ω
This proves (b).
Theorem 4.3 Let s : Ω × Ω → Cn satisfy the conditions (A), (B) and 1 ¯ = 0, then (C), and q > 0. If f ∈ C(p,q) (Ω) satisfies the equation ∂f p+q
u(z) = (−1)
Cp,q,n
Z
f (ζ) ∧ Kp,q−1 (z, ζ)
Ω
¯ = f. is a solution of the equation ∂u Proof. The condition (C) implies that ∂¯z sj = 0 for j = 1, · · · , n and ζ ∈ ∂Ω, and hence ∂¯z Qj = 0 for j = 1, · · · , n and ζ ∈ Ω. Since Qj are of class C 1 in Ω × Ω, we have ∂¯z Qj = 0 for j = 1, · · · , n and ζ ∈ ∂Ω. It follows from the condition (C) that N is of degree 0 with respect to d¯ z for ζ ∈ ∂Ω, which means that Kp,q = 0 for q > 0 by (4.10). Then Theorem 4.3 follows from Theorem 4.2 (a). Definition 4.1
For a = (a1 , · · · , an ) ∈ Cn , define ω ′ (a, ξ) =
n X j=1
(−1)j−1 aj ∧ dξj . i6=j
We have the following lemma. We omit the proof.
Several Complex Variables and Integral Formulas
254
Lemma 4.1 ω ′ (a, ξ) ∧ ω(η) = Cn
n X
k=1
ak dηk ∧
n X j=1
where Cn = (−1)n(n−1)/2 /(n − 1)!.
n−1
dξj ∧ dηj
,
For s = (s1 , · · · , sn ), Q = (Q1 , · · · , Qn ), define
Definition 4.2 s=
n X
sj (dζj − dzj ),
Q=
n X
Qj (dζj − dzj ).
j=1
j=1
Notice that we use notations which have two meanings. By Lemma 4.1 we have Nt = exp(< Q, ζ − z > +t < s, ζ − z >)dt ∧ ω ′ (s, Q + ts) ∧ ω(ζ − z) = Cn exp(< Q, ζ − z > +t < s, ζ − z >)dt ∧ s ∧ (dQ + tds)n−1 = Cn exp(< Q, ζ − z > +t < s, ζ − z >)dt ∧ s ∧ n−1 X n − 1 (dQ)k ∧ (ds)n−1−k tn−k−1 . k k=0
It follows from the definition of K that K = Cn (−1)n exp < Q, ζ − z >
n−1 X
(−1)k
k=0
(n − 1)! s ∧ (dQ)k ∧ (ds)n−1−k k! < s, ζ − z >n−k (4.13)
and P =
(−1)n(n−1)/2 exp < Q, ζ − z > (dQ)n . n!
(4.14)
For a C 1 function ψ : Ω × Ω → C\{0}, we have ψs ∧ (d(ψs))j = ψs ∧ (dψ ∧ s + ψds)j = ψ j+1 s ∧ (ds)j . Hence we may assume in (4.13) that Re < s, ζ − z >< 0. Theorem 4.4 (Berndtsson-Andersson formula) Assume that s satisfies the conditions A and B. Let a function G be holomorphic in a simply connected domain which contains {< Q(z, ζ), z − ζ > +1 | (ζ, z) ∈ Ω × Ω} and G(1) = 1. Define e = Cn (−1)n K
n−1 X k=0
(n − 1)! (k) s ∧ (dQ)k ∧ (ds)n−1−k G (< Q, z − ζ > +1) k! < s, ζ − z >n−k
Integral Formulas with Weight Factors
255
and (−1)n(n−1)/2 (n) P = G (< Q, z − ζ > +1)(dQ)n . n!
1 Then for f ∈ C(p,q) (Ω) one has (a) In case q > 0, ¯ ∧K p,q−1 p,q + (−1)p+q+1 p,q − ∂¯z f ∧K f ∧K ∂f , f =C ∂Ω
Ω
Ω
p,q are components of K which are (p, q) forms with respect to z where K and (n − p, n − q − 1) forms with respect to ζ. (b) In case q = 0, p+1 ¯ ∂f ∧ Kp,0 − f ∧ Kp,0 + (−1) f ∧ Pp,0 . f =C ∂Ω
Proof. Let
Ω
Ω
First we prove Theorem 4.4 in the case when G is a polynomial.
G(α) =
N
j
aj α ,
g=
j=0
N j=0
aj
dj δ , dλj
where δ is the Dirac delta function. We denote by K (λ) instead of K when we replace in (4.13) Q by λQ. Similarly, we denote by P (λ) instead of P when we replace in (4.14) Q by λQ. After replacing in (4.13) and (4.14), if we multiply by e−λ and operate g, then we obtain the desired equalities, where we have used the equations g(e−λ ) =
N j=0
G(α) = g(e−αλ ),
N
aj
dj δ −λ aj = G(1) = 1, (e ) = j dλ j=0
G(k) (< Q, z − ζ > +1) = g((−λ)k e−λ(+1) ).
In the general case, G is approximated uniformly in {< Q(z, ζ), z − ζ > +1 | (ζ, z) ∈ Ω × Ω} by a sequence of polynomials. 4.2
Lp Estimates for the ∂¯ Problem
Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C 2 boundary. In this section we prove Lp estimates for the ∂¯ problem in Ω. Lp estimates for the ∂¯ problem in Ω were first proved by Ovrelid [OV] and Kerzman [KER]
Several Complex Variables and Integral Formulas
256
using the homotopy formula discussed in Chapter 3. The proof given here is due to Bruna-Cufi-Verdera [BRV] using the Berndtsson-Andersson formula. Let ρ be a C 2 function in a neighborhood U of Ω such that Ω = {z ∈ U | ρ(z) < 0}, dρ(z) = 0 for z ∈ ∂Ω. For ε and δ > 0, define Vδ = {z ∈ U | |ρ(z)| < δ},
Ωδ = {z ∈ U | ρ(z) < δ},
Uε,δ = {(z, ζ) ∈ Ωδ × Vδ | |ζ − z| < ε}. We choose δ > 0 sufficiently small such that Vδ ⊂⊂ U . There exist β > 0 and ajk ∈ C 1 (V δ ) such that n ∂ 2 ρ(ζ) ¯ 2 inf ¯k ξj ξk ≥ 3β|ξ| ∂ζ ∂ ζ ζ∈V δ j j,k=1
(0 = ξ ∈ Cn ),
2 ∂ ρ(ζ) β − ajk (ζ) < 2 . sup ∂ζ ∂ζ n j k ζ∈V δ
For any sufficiently small ε > 0, if we set ζj = xj +ixn+j for j = 1, · · · , n, then we have 2 ∂ ρ(ζ) β ∂ 2 ρ(z) < − (ζ, z ∈ V δ , |ζ − z| < 2ε). ∂xj ∂xk ∂xj ∂xk 2n2
Instead of the Levi polynomial, we define F (z, ζ) by F (z, ζ) =
n ∂ρ(ζ) j=1
∂ζj
(ζj − zj ) −
n 1 ajk (ζ)(ζj − zj )(ζk − zk ). 2 j,k=1
Using Taylor’s theorem, we have 2Re F (z, ζ) ≥ ρ(ζ) − ρ(z) + β|ζ − z|2
(4.15)
for ζ, z ∈ V δ and |ζ − z| < 2ε. Moreover, using the same method as the proof of Theorem 3.8, we obtain the following lemma. Lemma 4.2 There exist constants ε, δ, c > 0 and functions Φ ∈ C 1 (Ωδ × Vδ ), G ∈ C 1 (Uε,δ ) with the following properties: (a) Φ(z, ζ) and G(z, ζ) are holomorphic in z for fixed ζ. (b) Φ = F G in Uε,δ . (c) |G| > c in Uε,δ , |Φ| > c in Ωδ × Vδ \Uε,δ .
Integral Formulas with Weight Factors
257
(d) There exist wj ∈ C 1 (Ωδ × Vδ ) for j = 1, · · · , n such that Φ(z, ζ) =
n j=1
wj (z, ζ)(ζj − zj ).
Moreover, wj (z, ζ), 1 ≤ j ≤ n, are holomorphic with respect to z. We set w(z, ζ) = (w1 (z, ζ), . . . , wn (z, ζ)). Let ϕ ∈ C ∞ (Cn ) be a function with the properties that 0 ≤ ϕ ≤ 1, ϕ = 1 in a neighborhood of ∂Ω, ϕ = 0 outside of Vδ . Define u(z, ζ) = −ρ(ζ)(ζ¯ − z¯) + w(z, ζ)Φ(z, ζ)ϕ(ζ)
((z, ζ) ∈ Ωδ × Ωδ ).
Since u(z, ζ) = w(z, ζ)Φ(z, ζ) for ζ ∈ ∂Ω, u is the product of w and a function. Hence u satisfies the condition (C) in Corollary 4.1. For (z, ζ) ∈ Ω × Ω we have < u(z, ζ), ζ − z >= −ρ(ζ)|ζ − z|2 + |Φ(z, ζ)|2 ϕ(ζ). Hence for ζ ∈ Ω with ζ = z, we have < u(z, ζ), ζ − z >≥ −ρ(ζ)|ζ − z|2 > 0. For ζ ∈ ∂Ω with ζ = z and |ζ − z| < ε, (4.15) shows that there exists c0 > 0 such that |Φ(z, ζ)| ≥ c0 . For ζ ∈ ∂Ω with |ζ − z| < ε, (c) shows that |Φ(z, ζ)| ≥ c. Since ϕ(ζ) = 1 for ζ ∈ ∂Ω, u satisfies the condition (A) in Theorem 4.1. Let K ⊂ Ω be a compact set. For z ∈ K with |ζ − z| ≤ 2ε, if ζ is contained in a small neighborhood B of ∂Ω, then ϕ(ζ) = 1 and ρ(ζ) − ρ(z) > 0. Then (4.15) shows that 2ReF (z, ζ) > β|ζ − z|2 . Hence there exists c > 0 such that |Φ(z, ζ)| > c|ζ − z|2 . Therefore, for ζ ∈ Ω ∩ B and |ζ − z| ≤ ε, there exists c′ > 0 such that < u(z, ζ), ζ − z >≥ −ρ(ζ)|ζ − z|2 + c′ |ζ − z|2 ≥ c′ |ζ − z|2 . For ζ ∈ Ω\B and |ζ − z| ≤ ε, there exists c1 > 0 such that −ρ(ζ) > c1 > 0, and hence < u(z, ζ), ζ − z >≥ c1 |ζ − z|2 . If |ζ − z| ≥ ε and ζ ∈ Ω, then there exists c′1 > 0 such that |Φ(z, ζ)| > c′1 . Hence for ζ ∈ Ω and z ∈ K, there exists c′′1 > 0 such that < u(z, ζ), ζ − z >≥ c′′1 |ζ − z|2 . Thus, u satisfies the condition (B) in Theorem 4.1. Next we modify u near the diagonal ∆ of ∂Ω × ∂Ω. Define Vε,δ := {(ζ, z) | |ρ(z)| < δ, |ρ(ζ)| < δ, |ζ − z| < ε}, a(z, ζ) := −ρ(ζ) + F (z, ζ),
Wε,δ := Vε,δ ∩ (Ω × Ω),
a∗ (z, ζ) := a(ζ, z).
258
Several Complex Variables and Integral Formulas
Further, for (z, ζ) ∈ V2ε,δ we define v(z, ζ) := ρ(ζ)
w(ζ, z) w(z, ζ) + a∗ (z, ζ) . G(ζ, z) G(z, ζ)
Since v(ζ, ζ) = 0, we have |v(z, ζ)| ≤ C|z − ζ|. On the other hand, we obtain < v, ζ − z >= −ρ(ζ)F (ζ, z) + +a∗ (z, ζ)F (z, ζ) = a(z, ζ)a∗ (z, ζ) − ρ(ζ)ρ(z). It follows from (4.15) that 2Re a(z, ζ) ≥ −ρ(ζ) − ρ(z) + β|ζ − z|2
(|ζ − z| < 2ε).
(4.16)
Consequently, for (z, ζ) ∈ W2ε,δ , there exists a constant c2 > 0 such that | < v, ζ − z > | ≥ |a||a∗ | − ρ(ζ)ρ(z)
≥ |Re a||Re a∗ | + |Im a||Im a∗ | − ρ(ζ)ρ(z)
≥ c2 {(ρ(ζ) − ρ(z))2 + |ζ − z|4 + (−ρ(ζ) − ρ(z))|ζ − z|2 +|Re F (z, ζ)||Re F (ζ, z)|}.
Hence if (z, ζ) ∈ W2ε,δ , then | < v, ζ − z > | ≥ (−ρ(ζ) − ρ(z))|ζ − z|2 , which implies that v satisfies (B). Since < v, ζ − z >= 0 for ζ = z, multiplying by < v, ζ − z >/| < v, ζ − z > |, we may assume that < v, ζ − z >> 0 in W2ε,δ \∆. Let λ ∈ C ∞ (Cn × Cn ) be a function such that 0 ≤ λ ≤ 1, λ = 1 in Vε,δ′ , where δ ′ < δ, λ = 0 outside of V2ε,δ . Define s = λv + (1 − λ)u. Then s : Ωδ × Ωδ → Cn is of class C 1 , < u, ζ − z >> 0 for ζ = z, and hence < s, ζ − z >> 0 for ζ = z. Therefore s satisfies (A). For ζ ∈ ∂Ω, s is a product of w(z, ζ) and a function. Thus s satisfies (C). Clearly s satisfies (B). By Theorem 4.2, if we set ψ = (s, ζ − z), K = ψ ∗ µ, then for 1 ¯ = 0, (Ω) with ∂f f ∈ C(p,q) T f (z) = Cp,q,n
Ω
f (ζ) ∧ Kp,q−1 (z, ζ)
¯ f ) = f . For ψ˜ = (v, ζ − z), we set K(v) = ψ˜∗ µ. Define satisfies ∂(T Fj (z, ζ) =
Pj (z, ζ) G(z, ζ)
(j = 1, · · · , n)
Integral Formulas with Weight Factors
259
and α(z, ζ) =
n j=1
Fj (z, ζ)(dζj − dzj ),
β(z, ζ) =
n j=1
Fj (ζ, z)(dζj − dzj ).
Then we have v=
n j=1
vj (dζj − dzj ) = ρ(ζ)β(z, ζ) + a∗ (z, ζ)α(z, ζ).
Consequently, dv = dρ ∧ β + ρdβ + da∗ ∧ α + a∗ dα. Since we can adopt the binomial theorem for 2-forms, it follows from Lemma 4.1 that k(v) =
n Cp,q,n (−1)j−1 vj ∧ dvi ∧ ω(ζ − z) i=j < v, ζ − z >n j=1
(−1)n(n−1)/2 Cp,q,n v ∧ (dv)n−1 (n − 1)! < v, ζ − z >n / −n = Cn < v, ζ − z > (ρβ + a∗ α) ∧ (ρdβ + a∗ dα)n−1
=
0 +(n − 1)(ρdβ + a∗ dα)n−2 ∧ ((−ρ)da∗ + a∗ dρ) ∧ β ∧ α ,
where Cn is a constant such that Cn =
(−1)n(n−1)/2 Cp,q,n . (n − 1)!
It follows from (4.16) that −ρ(ζ) ≤ 2|a∗ (z, ζ)|
((z, ζ) ∈ Wε,δ ).
Since ρ(ζ)β + a∗ α =
n j=1
{ρ(ζ)Fj (ζ, z) − ρ(z)Fj (z, ζ) + F (ζ, z)Fj (z, ζ)}(dζj − dzj )
and β∧α=
j |n
((z, ζ) ∈ Wε,δ ).
For (z, ζ) ∈ Vε,δ , define T (z, ζ) = |F (z, ζ)| + |F (ζ, z)|. Then we have the following lemma. Lemma 4.3 (a) (b) (c) (d) (e)
Let (z, ζ) ∈ Wε,δ . Then
a(z, ζ) ≈ a∗ (z, ζ) ≈ −ρ(ζ) − ρ(z) + |ζ − z|2 + |Im F (z, ζ)|, T (z, ζ) ≈ |ρ(ζ) − ρ(z)| + |ζ − z|2 + |Im F (z, ζ)|, | < v, ζ − z > | ≥ C{T (z, ζ)2 + (−ρ(ζ) − ρ(z))|ζ − z|2 }, |ζ − z|2 |a∗ (z, ζ)| ≤ C| < v, ζ − z > | ≤ |a∗ (z, ζ)|2 , | < v, ζ − z > | ≤ C|ζ − z||a∗ (z, ζ)|,
where C is a constant which is independent of ζ and z. Proof.
It follows from Taylor’s formula that
ρ(z) = ρ(ζ) n n 2 ∂ρ ∂ ρ +Re 2 (ζ)(zj − ζj ) + (ζ)(zj − ζj )(zk − ζk ) ∂z ∂z ∂z j j k j=1 j,k=1
+
n
j,k=1
2
∂ ρ (ζ)(zj − ζj )(¯ zk − ζ¯k ) + O(|ζ − z|3 ) ∂zj ∂ z¯k
= ρ(ζ) − 2Re F (z, ζ) n ∂2ρ + (ζ)(zj − ζj )(¯ zk − ζ¯k ) + O(|ζ − z|3 ). ∂zj ∂ z¯k j,k=1
Consequently, 2Re F (z, ζ) ≤ ρ(ζ) − ρ(z) + C|ζ − z|2 . Hence together with (4.16) we obtain a(z, ζ) ≈ −ρ(ζ) − ρ(z) + |ζ − z|2 + |Im F (z, ζ)|. Since |F (z, ζ) + F (ζ, z)| = O(|ζ − z|2 ),
Integral Formulas with Weight Factors
261
we have a∗ (z, ζ) ≈ −ρ(z) − ρ(ζ) + |ζ − z|2 + |ImF (ζ, z)|
≤ C(−ρ(z) − ρ(ζ) + |ζ − z|2 + |ImF (z, ζ)|) ≈ a(z, ζ).
This proves (a). If ρ(ζ) ≥ ρ(z), then |F (z, ζ)| + |F (ζ, z)|
≈ |Re F (z, ζ)| + |Im F (z, ζ)| + |Re F (ζ, z)| + |Im F (ζ, z)|
≥ |Re F (z, ζ)| + |Im F (z, ζ)|
≥ C(|ρ(ζ) − ρ(z)| + |ζ − z|2 + |ImF (z, ζ)|). We can also prove the above inequality in case ρ(ζ) ≤ ρ(z). Since |F (z, ζ) + F (ζ, z)| = O(|ζ − z|2 ), we obtain |F (z, ζ)| + |F (ζ, z)| ≤ 2|F (z, ζ)| + O(|ζ − z|2 )
≤ C(|ρ(ζ) − ρ(z)| + |ζ − z|2 + |ImF (z, ζ)|).
This proves (b). If |Im F (ζ, z)| ≥ |Im F (z, ζ)|, then | < v, ζ − z > |
≥ C{(ρ(ζ) − ρ(z))2
+|ζ − z|4 + |Im F (z, ζ)|2 + (−ρ(ζ) − ρ(z))|ζ − z|2 }
≥ C{T (z, ζ)2 + (−ρ(ζ) − ρ(z))|ζ − z|2 }.
We can also prove the above inequality in case |Im F (ζ, z)| ≤ |Im F (z, ζ)|. This proves (c). From the definition of a∗ and T we have |a∗ | ≤ −ρ(z) + |F | ≤ −ρ(z) + T. By (b) and (c) we have |ζ − z|2 |a∗ | ≤ −ρ(z)|ζ − z|2 + |ζ − z|2 T
≤ |ζ − z|2 (−ρ(z) − ρ(ζ)) + CT 2
≤ C| < v, ζ − z > |. It follows from (a) that
| < v, ζ − z > | = |aa∗ − ρ(ζ)ρ(z)| ≤ C|a∗ |2 .
Several Complex Variables and Integral Formulas
262
This proves (d). We obtain | < v, ζ − z > | = |aa∗ − ρ(z)ρ(ζ)|
= |aa∗ + ρ(ζ)a∗ − ρ(ζ)a∗ − ρ(z)ρ(ζ)| ≤ |a + ρ(ζ)||a∗ | − ρ(ζ)|a∗ + ρ(z)|
≤ C(|a + ρ(ζ)||a∗ | + |a∗ ||a∗ + ρ(z)|)
= C(|F (z, ζ)||a∗ | + |a∗ ||F (ζ, z)) ≤ C|ζ − z||a∗ |.
This proves (e). Lemma 4.4
There exists a constant C > 0 such that
B(z,r)∩Ω
|Kp,q (z, ζ)|dV (ζ) ≤ Cr
(z ∈ Ω, r > 0).
Proof. It is sufficient to prove the lemma under the assumption that r > 0 is sufficiently small and z is sufficiently close to ∂Ω. For (z, ζ) ∈ Wε,δ , we obtain K = K(v). By the definition of a∗ and T and using (c) we have ∗ |a (z, ζ)|n−1 |ζ − z| |Kp,q (z, ζ)| ≤ C | < v, ζ − z > |n (|ρ(z)|n−1 + |F (z, ζ)|n−1 )|ζ − z| ≤C [T (z, ζ)2 + (−ρ(ζ) − ρ(z))|ζ − z|2 ]n |ζ − z| |ρ(z)|n−1 |ζ − z| ≤C . + T (z, ζ)n+1 [T (z, ζ)2 + |ρ(z)||ζ − z|2 ]n We choose a coordinate system η1 (ζ), · · · , ηn (ζ) in a neighborhood of z such that η1 (ζ) = ρ(ζ) − ρ(z) + iIm F (z, ζ),
η(z) = 0,
|η(ζ)| ≈ |ζ − z|.
We set ηj = t2j−1 + it2j . Then we have |T | ≈ |t1 | + |t2 | + |t|2 ,
|ζ − z| ≈ |t|.
In order to prove Lemma 4.4, it is sufficient to show that |t| dt · · · dt2n ≤ Cr I1 = 2 n+1 1 |t|≤r [|t1 | + |t2 | + |t| ] and I2 =
2 |t|≤r [t1
|ρ(z)|n−1 |t| dt1 · · · dt2n ≤ Cr. + t22 + |ρ(z)||t|2 ]n
(4.17)
(4.18)
Integral Formulas with Weight Factors
263
We set t′ = (t3 , · · · , t2n ). Then we obtain dt1 · · · dt2n I1 ≤ [|t | + |t2 | + |t|2 ]n+(1/2) 1 |t|≤r dt′ ≤C ′ 2n−3 |t′ |≤r |t | r ds = Cr. ≤C 0
This proves (4.17). We set m = |ρ(z)|. Then we have
mn−(3/2) dt · · · dt2n 2 2 2 n−(1/2) 1 |t|≤r [t1 + t2 + m|t| ] r ds ≤ Cmn−(3/2) dt′ ′ |2 ]n−(1/2) ′ [s + m|t |t |≤r 0 r λ2n−3 ≤ Cmn−(3/2) dλ 2 n−(3/2) 0 [mλ ] r dλ = Cr. =C
I2 ≤
0
This proves (4.18).
Now we are going to prove Lp estimates for the ∂¯ problem in a strictly pseudoconvex domain Ω in Cn with smooth boundary. Theorem 4.5
1 For f ∈ C(p,q) (Ω), define
T f (z) = Cp,q,n
Ω
f (ζ) ∧ Kp,q−1 (z, ζ).
Then T satisfies the following: ¯ = 0, then ∂(T ¯ f) = f. (a) If ∂f r (b) If f ∈ L(p,q) (Ω) and 1 ≤ r ≤ ∞, then T f ∈ Lr(p,q) (Ω). Proof. Since every Lp function in Ω can be approximated uniformly on every compact subset of Ω by functions in C 1 (Ω), we may assume that f ∈ C 1 (Ω). (a) follows from Corollary 4.1. (b) follows from Lemma 4.4 and Theorem 3.26. Bruna-Cufi-Verdera [BRV] proved the following theorem. We omit the proof.
Several Complex Variables and Integral Formulas
264
Theorem 4.6 Let Ω ⊂⊂ Cn be a strictly pseudoconvex domain with C 3 boundary. Then each function f ∈ C(D) satisfying ∂¯i ∂¯j f = 0
(1 ≤ i, j ≤ n)
(4.19)
in Ω, can be approximated on Ω by functions satisfying (4.19) in a neighborhood of Ω.
4.3
The Berndtsson Formula
We study the integral formula on submanifolds of bounded domains with smooth boundary obtained by Berndtsson [BR1]. Let Ω be a bounded domain in Cn with C 2 boundary. Let Ω = {z | ρ(z) < 0}, where ρ is a C 2 function in a neighborhood of Ω and dρ = 0 on ∂Ω. Let h1 , · · · , hm be holomorphic functions in a neighborhood of Ω satisfying Ω ∂h1 ∧ · · · ∧ ∂hm ∧ ∂ρ = 0
(4.20)
on ∂Ω. Suppose there exist holomorphic functions gij (z, ζ) in Ω × Ω such that hj (z) − hj (ζ) =
n i=1
gij (z, ζ)(zi − ζi )
(j = 1, · · · , m).
(4.21)
We set | h1 (z) = · · · = hm (z) = 0}, X = {z ∈ Ω V = X ∩ Ω,
h = (h1 , · · · , hm ), j
g =
n
gij dζi ,
(4.22)
i=1
µ=
g 1 ∧ · · · ∧ g m ∧ ∂h1 ∧ · · · ∧ ∂hm dVn−1 , ∂h2
(4.23)
Integral Formulas with Weight Factors
265
where dVn−1 is the surface measure on V . Let s = (s1 , · · · , sn ), Q = (Q1 , · · · , Qn ) and G denote the same notations as in Theorem 4.4. Moreover, we use the abbreviation s=
n
sj dζj ,
n
Q=
Qj dζj .
j=1
j=1
We set K=
n−m−1 k=0
¯ n−m−1−k ∧ (∂Q) ¯ k∧µ (n − 1)! (k) s ∧ (∂s) , G (< Q, z − ζ > +1) n−m−k m!k! < s, ζ − z >
and P =
1 ¯ n−m ∧ µ. G(n−m) (< Q, z − ζ > +1)(∂Q) m!(n − m)!
Then we have the following theorem. Theorem 4.7 (Berndtsson formula) Let u be a C 1 (0, q) form on V . For z ∈ V one has (a) In case q > 0, u(z) = C
∂V
u ∧ Kq + (−1)q+1
V
¯ ∧ Kq − ∂¯z ∂u
V
u ∧ Kq
,
where Kq is a component of K which is of degree (0, q) with respect to z and of degree (n − m, n − m − q − 1) with respect to ζ and C = Cq,n is a constant depending only on q, n. (b) In case q = 0, ¯ u(z) = Cn uK0 − ∂u ∧ K0 − uP0 . ∂V
V
V
Proof. We prove Theorem 4.7 in case m = 1. Let h1 = h, g 1 = g. Suppose Q1 (z, ζ) and Q2 (z, ζ) are of class C 1 in Ω × Ω, and holomorphic in z ∈ Ω. In (4.14) we replace Q by λ1 Q1 + λ2 Q2 and P by P λ . Then we have (−1)n(n−2)/2 λ1 λ2 e e (λ1 dQ1 + λ2 dQ2 )n n! 1 2 = (−1)n(n−1)/2 eλ1 eλ2 λα1 λα2 1 2 (dQ1 )α1 (dQ2 )α2 . × α !α ! 1 2 α +α =n
Pλ =
1
2
Several Complex Variables and Integral Formulas
266
Suppose ψ1 , ψ2 are distributions. We set ∞ ∞ P = P λ e−λ1 e−λ2 ψ1 (λ1 )ψ2 (λ2 )dλ1 dλ2 . 0
0
Further we set G1 (α) =
∞
e−αλ1 ψ1 (λ1 )dλ1 ,
G2 (α) =
∞
e−αλ2 ψ2 (λ2 )dλ2 .
0
0
Then we have Q , z − ζ > +1) =
(< Q2 , z − ζ > +1) =
(α ) G1 1 (
+1). Since u is a (0, q) form, we have only to consider the terms in dQ1 and dQ2 which do not contain dzj , and ¯ 1 and ∂Q ¯ 2 , respectively. In (4.13) hence we may replace dQ1 and dQ2 by ∂Q 1 2 λ we replace Q by λ1 Q + λ2 Q and K by K . We set ∞ ∞ = K K λ e−λ1 e−λ2 ψ1 (λ1 )ψ2 (λ2 )dλ1 dλ2 . 0
0
Then we have
= Cn (−1)n × K (n − 1)!
α0 +α1 +α2 =n−1
α1 !α2 !
(α1 )
G1
(α2 ) s
G2
¯ 1 )α1 ∧ (∂Q ¯ 2 )α2 ¯ α0 ∧ (∂Q ∧ (∂s) . α +1 < s, ζ − z > 0
We choose distributions ψ1 , ψ2 such that G1 (1) = G2 (1) = 1. Using the same method as the proof of Theorem 4.4, we may assume that Gj are holomorphic in some simply connected domain containing {< Qj , z − ζ > +1 | (z, ζ) ∈ Ω × Ω}. Let g = (g1 , · · · , gn ). We set Q2ε =
h(ζ)g . |h(ζ)|2 + ε
Integral Formulas with Weight Factors
267
It follows from (4.21) that < Q2ε , z − ζ > +1 = = =
n h(ζ)gj (zj − ζj ) +1 |h|2 + ε j=1
h(ζ)(h(z) − h(ζ)) +1 |h|2 + ε h(ζ)h(z) + ε . |h|2 + ε
On the other hand we have Q2ε =
n h(ζ)gj (z, ζ) j=1
|h|2 + ε
dζj .
Consequently, ¯ 2 ∂Q ε
n n ∂ = ∂ ζ¯k j=1 k=1
where g =
n
j=1
h(ζ)gj |h|2 + ε
dζ¯k ∧ dζj =
ε∂h ∧ g , (|h|2 + ε)2
gj dζj . Therefore we have ¯ 2 )p = 0 (∂Q ε
(p > 1).
For simplicity, we assume h(z) = z1 . Then we have the following lemma. The proof is the same as the proof of Lemma 2.32. So we omit the proof. Lemma 4.5 lim ε→0+
lim
ε→0+
Ω
∂Ω
Let z = (z1 , z ′ ). For ϕ ∈ C 1 (Ω) we have ε ϕ(z)dVn−1 (z ′ ), ϕ(z)dV (z) = π (|z1 |2 + ε)2 {z1 =0}∩Ω ε ϕ(z)dσ2n−1 (z) = π (|z1 |2 + ε)2
(4.24)
ϕ(z)dσ2n−3 (z ′ ),
{z1 =0}∩∂Ω
(4.25) where dσ2n−1 and dσ2n−3 are surface measures on ∂Ω and {zn = 0} ∩ ∂Ω, respectively. We set G2 (α) = α. Then = (−1)n Cn K
α0 +α1 +α2
(n − 1)! (α1 ) G α1 !α2 ! 1 =n−1
h(ζ)h(z) + ε |h|2 + ε
1−α2
(4.26)
Several Complex Variables and Integral Formulas
268
×
¯ 2 )α2 ¯ 1 )α1 ∧ (∂Q ¯ α0 ∧ (∂Q s ∧ (∂s) ε < s, ζ − z >α0 +1
and Pe = (−1)
n(n+1)/2
X
α1 +α2
1 (α ) G1 1 α !α ! 1 2 =n
h(ζ)h(z) + ε |h|2 + ε
!1−α2
(4.27)
¯ 2 )α2 . ¯ 1 )α1 ∧ (∂Q ×(∂Q e and Pe are bounded by integrable Let α2 = 1. Then coefficients of K functions which are independent of ε. Let α2 = 0. Then we have
|h(ζ)h(z) + ε| ksk 1 |h(ζ)||ζ − z| ≤C 1+ . |h|2 + ε | < s, ζ − z > |n |h|2 + ε |ζ − z|2n−1 (4.28) In case |ζ − z| ≤ |h(ζ)|, the right side of (4.28) is bounded by |ζ − z|−2n+1 . In case |ζ − z| ≥ |h(ζ)|, if 0 < δ < 12 , ζ ′ = (ζ2 , · · · , ζn ), then there exist positive constants C1 , C2 and C3 such that Z ksk |h(ζ)||ζ − z| dV (ζ) 2+ε |h| | < s, ζ − z > |n Ω Z |h(ζ)|δ |ζ − z|1+δ ≤ C1 dV (ζ) (|h|2 + ε)|ζ − z|2n−1 ZΩ Z dV (ζ1 ) dVn−1 (ζ ′ ) ≤ C2 . 2−δ ′ ′ 2n−2−δ |ζ1 | +1 =
< ∂ρ(ζ), z − ζ > +ρ(ζ) − ε , ρ(ζ) − ε
It follows from (4.31) that Re(< Q, z − ζ > +1) ≥
ρ(z) + ρ(ζ) − 2ε >0 2(ρ(ζ) − ε)
for (z, ζ) ∈ Ω × Ω. We set G(α) = α−N for N ≥ 1. Since G(α) is holomorphic in Re α > 0, G satisfies the hypothesis in Theorem 4.7. If we let ε ↓ 0, then by Theorem 4.7 we have for some constant γk , N +k n−m−1 ¯ n−m−1−k ∧ (∂Q) ¯ k ∧µ ρ s ∧ (∂s) , K= γk < ∂ρ, z − ζ > +ρ < s, ζ − z >n−m−k k=0 (4.32)
Several Complex Variables and Integral Formulas
270
P = Cn,m
ρ(ζ) < ∂ρ(ζ), z − ζ > +ρ(ζ)
N +n−m
¯ n−m ∧ µ. (∂Q)
(4.33)
Since n n ∂ρ ∂ρ 1 ¯ ¯ = 1 ∂¯ (ζ)dζj − 2 ∂ρ (ζ)dζj , ∂Q ∧ ρ ∂ζ ρ ∂ζ j j j=1 j=1
¯ k = O(|ρ|−k−1 ). On the other hand we have we obtain (∂Q)
2(Re < ∂ρ, z − ζ > +ρ(ζ)) < ρ(z) + ρ(ζ) ≤ ρ(z) (z ∈ Ω, ζ ∈ Ω), which implies that the integral of K on Ω exists. Since dρ = 0 on ∂Ω, we have n ∂ρ 1 ¯ = ∂¯ dζj , ∂Q ρ ∂ζ j j=1
¯ k = O(|ρ|−k ). Since N ≥ 1, the integral of K on ∂Ω is and hence (∂Q) equal to 0. Thus by Theorem 4.7 (b) we obtain (4.29).
Remark 4.1 In the case when Ω is an analytic polyhedron, Lp and H p extensions of holomorphic functions from submanifolds of Ω were investigated by Adachi-Andersson-Cho [ADC] using the Berndtsson integral formula. 4.4
Counterexamples for Lp (p > 2) Extensions
We give counterexamples for Lp (p > 2) extensions of holomorphic functions from submanifolds in complex ellipsoids due to Mazzilli [MAZ1] and Diederich-Mazzilli [DIM1]. From these examples one can see that the Ohsawa-Takegoshi extension theorem is the best possible. Let Ω be a complex ellipsoid in Cn . Then there exist positive integers q1 , · · · , qn such that Ω = {z ∈ Cn | ρ(z) =
n j=1
|zj |2qj − 1 < 0}.
We set k = sup {qj }, 1≤j≤n
Integral Formulas with Weight Factors
and Q(z, ζ) =
271
∂ρ ∂ρ (ζ), · · · , (ζ) . ∂ζ1 ∂ζn
The following two lemmas have been proved by Range [RAN1]. Lemma 4.6 For a positive integer m , we set g(z) = |z|2m . Then there exists a constant C > 0 such that ∂g g(z + w) − g(z) − 2Re (z)w ≥ C|w|2m . ∂z Proof.
We set f (z, w) = g(z + w) − g(z) − 2Re
∂g (z)w . ∂z
By Taylor’s formula, if we set z = x + iy, w = u + iv, then there exists θ with 0 < θ < 1 such that 1 ∂2g ∂2g 2 (x + θu, y + θv)uv f (z, w) = (x + θu, y + θv)u + 2 2 ∂x2 ∂x∂y ∂ 2g + (x + θu, y + θv)v 2 . ∂y 2 We set ψ(t) = tm . Then we have g(z) = ψ(x2 + y 2 ). We set X = x + θu, Y = y + θv. Then f (z, w) = 2ψ ′′ (X 2 + Y 2 )(Xu + Y v)2 + ψ ′ (X 2 + Y 2 )(u2 + v 2 ).
(4.34)
Suppose (X, Y ) = (0, 0) for |w| = 1. Since z = −θw, we have |z| = θ. Consequently, f (z, w) = (1 − θ)2m − θ2m + 2mθ2m−1 = (1 − θ)2m + θ2m−1 (2m − θ) > 0. From (4.34) we have f (z, w) = 0, which is a contradiction. Hence we have (X, Y ) = 0, which means that f (z, w) > 0 for |w| = 1. In case |w| = 1 and |z| ≤ 2, f (z, w) has a minimum value c1 > 0. In case |w| = 1 and |z| ≥ 2, by (4.34) we have f (z, w) ≥ ψ ′ (X 2 + Y 2 ) = m(|z + θw|2 )m−1 ≥ m(|z| − 1)2m−2 ≥ m. Hence if |w| = 1, then f (z, w) ≥ min(c1 , m) := c. Since z w 1 f , f (z, w) ≥ c, = |w| |w| |w|2m
Several Complex Variables and Integral Formulas
272
we obtain f (z, w) ≥ c|w|2m . Lemma 4.7
For (ζ, z) ∈ Ω × Ω, there exists a constant C > 0 such that
Re(< Q, ζ − z > −ρ(ζ)) ≥ C(−ρ(ζ) − ρ(z) + 2qj
+|zj − ζj | Proof.
).
n j=1
|ζj |2qj −2 |zj − ζj |2
We set g(z) = |z|2m , ϕ(t) = g(z + tw) for t ∈ R. Then we have s ∂ ∂ ϕ(s) (t) = w +w ¯ g. ∂z ∂ z¯
In case 2m ≥ s, (s)
ϕ
(t) =
n j=0
=
s! j!(s − j)!
j+k=s
j s−j ∂ ∂ w ¯ g w ∂z ∂ z¯
s! ∂ n g wj w ¯k . j!k! ∂z j ∂ z¯k
In case 2m < s, we have ϕ(s) (t) = 0. Hence we have f (z, w) =
2m ϕ(s) (0) s=2
We obtain
j+k=2
s!
=
2≤j+k≤2m
∂sg 1 wj w ¯k . j!k! ∂z j ∂ z¯k
∂ sg 1 wj w ¯k j!k! ∂z j ∂ z¯k
= m2 |z|2m−2 |w|2 + Re(m(m − 1)z m−2 z¯m w2 )
≥ m2 |z|2m−2 |w|2 − m(m − 1)|z|2m−2 |w|2 = m|z|2m−2 |w|2 .
On the other hand, if j + k ≥ 3, then for a with 0 < a < 1 and |w| ≤ a|z|, there exists a constant C > 0 such that ∂ j ∂ k g(z)wj w ¯k ∂z ∂ z¯ = |m(m − 1) · · · (m − j + 1)z m−j m(m − 1) · · · (m − k + 1)¯ z m−k wj w ¯k | ≤ Ca|z|2m−2 |w|2 ,
which means that for any sufficiently small a, 2
f (z, w) ≥ m|z|2m−2 |w|2 − Ca|z|2m |w|2 ≥ C|z|2m−2 |w|2 .
Integral Formulas with Weight Factors
273
If |w| > a|z|, then |w|2m > a2m−2 |z|2m−2 |w|2 , and hence by Lemma 4.6 we have f (z, w) ≥ C|w|2m ≥ C|z|2m−2 |w|2 . Consequently, we obtain f (z, w) ≥ C(|z|2m−2 |w|2 + |w|2m ).
(4.35)
We set gj (zj ) = |zj |2qj . It follows from (4.35) that n n n ∂g j (ζj )(zj − ζj ) gj (ζj ) − 2Re gj (zj ) − ∂ζ j j=1 j=1 j=1 n n |zj − ζj |2mj . ≥C |ζj |2qj −2 |zj − ζj |2 + j=1
j=1
Then
2Re
≥ −ρ(z) + ρ(ζ) + C( |ζj |2qj −2 |zj − ζj |2 ∂ζ j=1
+
n j=1
Definition 4.3
|zj − ζj |2mj ).
For k > 0, define
Bk (Ω) = {f ∈ C(Ω) | sup(|f (z)|dist(z, ∂Ω)k ) < ∞}. z∈Ω
Now we give a counterexample for bounded extensions of holomorphic functions from submanifolds of complex ellipsoids due to Mazzilli [MAZ1]. Theorem 4.9 For p ≥ 1 and any sufficiently small ε > 0, there exist a complex ellipsoid Ω in C2p+1 , a submanifold X in a neighborhood of Ω which intersects ∂Ω transversally, and a bounded holomorphic function f in V = X ∩ Ω such that if g is a holomorphic function in Ω with g|V = f , then g ∈ B p2 −ε (Ω). Therefore, f cannot be extended to a bounded holomorphic function in Ω. Proof. Define fj (z) = zjn + zp+j for j = 1, · · · , p. Define Ω ⊂ C2p+1 and a submanifold M of Ω as follows. Ω = {z ∈ C
2p+1
|
p j=1
2n+1
|zj |
+
2p+1
j=p+1
|zj |2 − 1 = ρ(z) < 0},
274
Several Complex Variables and Integral Formulas
V = {z ∈ Ω | f1 (z) = · · · = fp (z) = 0}. In Lemma 4.7, we set ζ = (0, · · · , 0, 1). Then for z ∈ Ω we have 2p p n+1 |zj |2 , |zj |2 + Re (1 − z2p+1 ) ≥ C |ρ(z)| + |z2p+1 − 1|2 + j=p+1
j=1
(4.36) which implies that if z ∈ Ω, then Re (1 − z2p+1 ) > 0. Hence if we define for z∈Ω f (z) =
z1n−1 · · · zpn−1
(1 − z2p+1 )
p(n−1) 2n
,
then f is holomorphic in Ω. It follows from (4.36) that |f (z)| ≤
|z1n−1 · · · zpn−1 | |1 − z2p+1 |
p(n−1) 2n
n−1
n−1
|zp+1 | n · · · |z2p | n ≤ 2p ≤C p(n−1) ( j=p+1 |zj |2 ) 2n
for z ∈ V , which means that f is bounded on V . It follows from (4.36) that n+1
|zj |2 ≤C |1 − z2p+1 |
(j = 1, · · · , p)
for z ∈ Ω. Consequently, n−1 n−1 n+1 n+1 2n+1 2n+1 |z1 |2 |zp |2 |f (z)| = ··· |1 − z2p+1 | |1 − z2p+1 | ×|1 − z2p+1 | ≤ C|ρ(z)|
p(n−1) − p(n−1) 2n 2n+1
p(n−1) − p(n−1) 2n 2n+1
.
Suppose there exists a holomorphic function g in Ω with the following properties: (a) For z ∈ V , g(z) = f (z).
(b) There exists δ > 0 such that |g(z)| ≤ C|ρ(z)|−
p(n−1) +δ + p(n−1) 2n 2n+1
.
Since g − f is a holomorphic function in Ω such that g − f = 0 on V , it follows from Corollary 5.7 that there exist holomorphic functions ak for k = 1, · · · , p in Ω such that p 1 n−1 n n−1 (zk + zp+k )ak (z) . (4.37) z1 · · · zp + g(z) = p(n−1) (1 − z2p+1 ) 2n k=1
Integral Formulas with Weight Factors
275
For any sufficiently small ε > 0 and θk ∈ [0, 2π] for k = 1, · · · , p, we set 1 = ε 2n+1 eiθk (1 ≤ k ≤ p) zk =0 (p + 1 ≤ k ≤ 2p) . zk z = 1 − pε 2p+1
Then we have ρ(z) = pε(pε − 1) < 0, which means that |ρ(z)| = (1 − pε)pε for z ∈ Ω. It follows from (b) and (4.37) that p p(n−1) p(n−1) 1 ak (z) 1 1 . + ≤ Cn |ρ(z)|− 2n + 2n+1 +δ n p p(n−1) n n z z · · · z |z · 1 p 2n j=1 1 · · zp | (pε) j k=1
j=k
Since |ρ(z)| ≈ ε, we obtain p p 1 a (z) k ≤ Cn εδ− 2n+1 . p z1 · · · zp + n j=1 zj k=1 j=k
We set
1
γ = {z ∈ C | |z| = ε 2n+1 }, Γ = γ × ··· × γ. * +( ) p
Then we have
p a (z) pk n dz1 ∧ · · · ∧ dzp = 0. Γ j=1 zj k=1
Consequently,
Γ
j=k
p 1 a (z) 1 pk n dz1 ∧ · · · ∧ dzp . dz1 ∧ · · · ∧ dzp = + z1 · · · zp z · · · z 1 p Γ j=1 zj
k=1
j=k
The left side of the above equality is equal to (2πi)p and the right side is equal to O(εδ ), which is a contradiction for any sufficiently small ε. Therefore there is no g which satisfies (a) and (b). Suppose for an extension g of f there exists ε > 0 such that p
|g(z)||ρ(z)| 2 −ε < C
(z ∈ Ω).
Several Complex Variables and Integral Formulas
276
If we choose n sufficiently large, then we have p(n − 1) p ε + < . 2n 2n+1 2 Consequently, p
|g(z)| ≤ C|ρ(z)|− 2 +ε ≤ C|ρ(z)|−
p(n−1) + ε2 + p(n−1) 2n 2n+1
,
which means that g satisfies (a) and (b). This is a contradiction.
Definition 4.4 Suppose d(z, ∂Ω) denotes the distance from z to ∂Ω and that dVΩ and dVn−1 are Lebesgue measures on Ω and V , respectively. (1) For a measurable function f in Ω, f ∈ Lq (d(z, ∂Ω)s , dVΩ ) means that |f (z)|p d(z, ∂Ω)s dVΩ < ∞. Ω
(2) For a measurable function f in V , f ∈ Lq (d(z, ∂Ω)s , dVn−1 )) means that |f (z)|p d(z, ∂Ω)s dVn−1 < ∞. V
Lemma 4.8 Let n and p be positive integers with n ≥ 2p + 1. For a positive integer N ≥ 2, define n
Ω = {z ∈ C |
p j=1
N +1
|zj |2
+
n
j=p+1
|zj |2 − 1 = ρ(z) < 0}.
Let q ≥ 2, s ≥ 0. Then there exists a constant CN > 0 such that for a holomorphic function f in Ω with f ∈ Lq (d(z, ∂Ω)s , dVΩ ), θj ∈ [0, 2π] and any sufficiently small ε > 0, if we set 1
1
z = (ε 2N +1 eiθ1 , · · · , ε 2N +1 eiθp , 0, · · · , 0, 1 − pε), then we obtain − n−p+1 − qs − q
|f (z)| ≤ CN f Lq (d(z,∂Ω)s ,dVΩ ) d(z, ∂Ω) Proof.
p 2N q
.
We set
P (z, ζ) =
ρ(ζ) < ∂ρ(ζ), z − ζ > +ρ(ζ)
1+ qs +n n 1 ¯ ∧ µ, ∂ ∂ log − ρ(ζ)
Integral Formulas with Weight Factors
277
where µ is defined by (4.23). It follows from Theorem 4.4 that for f ∈ O(Ω) ∩ Lq (d(z, ∂Ω)s , dVΩ ) f (ζ)P (z, ζ). f (z) = Ω
By the H¨older inequality we obtain |f (z)| q−1 1q q q − sq q s ≤ |f (ζ)| |ρ(ζ)| dVΩ (ζ) . (|ρ(ζ)| |P (z, ζ)|) q−1 dVΩ (ζ) Ω
Ω
Further we have n ¯ ¯ n ¯ n−1 ∧ ∂ρ ∧ ∂ρ ∂ρ (∂∂ρ) (∂∂ρ) = −n . ∂¯ n n+1 ρ ρ ρ Consequently, s
|ρ(ζ)|− q |P (z, ζ)| ≤ C
n−1 ¯ ¯ (∂∂ρ(ζ)) ∧ ∂ρ(ζ) ∧ ∂ρ(ζ) . s | < ∂ρ(ζ), z − ζ > +ρ(ζ)|1+ q +n
It follows from Lemma 4.7 that Re < ∂ρ(ζ), ζ − z > −ρ(ζ) p n N +1 ≥ C(|ρ(ζ)| + |ρ(z)| + |zj − ζj |2 ). |zj − ζj |2 + j=p+1
j=1
We set s
P1 (z, ζ) = ρ(ζ)− q P (z, ζ) and un = ρ(ζ) + iIm < ∂ρ(ζ), z − ζ > . Then we have p
N +1
|ζj |2 −2 |P1 (z, ζ)| ≤ C p n−1 s . (|ρ(z)| + j=1 |ζj − zj |2N +1 + j=p+1 |ζj |2 + |un |)n+1+ q j=1
Since
1
1
z = (ε 2N +1 eiθ1 , · · · , ε 2N +1 eiθp , 0, · · · , 0, 1 − pε),
Several Complex Variables and Integral Formulas
278
we obtain p
|ρ(z)| = 1 −
N +1
|zj |2
j=1
N +1
which implies that |ρ(z)| ≈ |zj |2 we have p
j=1
N +1
|ζj |2
−2
≤C
p
j=1
− |zn |2 = pε(1 − pε),
for j = 1, · · · , p. On the other hand N +1
(|ζj − zj |2
N +1
−2
+ |zj |2
N +1
)1− 2N ,
−2
)
and N +1
|ζj − zj |2
−2
= (|ζj − zj |2
1
which means that |P1 (z, ζ)| ≤C We set
(|ρ(z)| +
p
j=1
|ζj − zj
α=−
|2N +1
+
1 n−1
j=p+1
s
s p q (n − p + 1 + + N ) q−1 q 2
and u = (u1 , · · · , un ),
u′ = (u1 , · · · , un−1 ),
u′′ = (u1 , · · · , up ).
Define uj = ζj − zj for j = 1, · · · , n − 1. Then we have q |P1 (z, ζ)| q−1 dλΩ (ζ) Ω du′ ≤C p 2 −α−2 (|ρ(z)| + j=1 |uj |2N +1 + n−1 j=p+1 |uj | ) du′′ ≤C p 1 (|ρ(z)| 2 + j=1 |uj |2N )−2α−4−2(n−p−1) du′′ ≤C 1 p (|ρ(z)| 2N + j=1 |uj |2 )−2N (α+n−p+1) R r2p−1 dr ≤C 1 N 0 (|ρ(z)| 2N + r2 )−2 (α+n−p+1) p
≤ C|ρ(z)| 2N +α+n−p+1 .
p
|ζj |2 + |un |)n+1−p+ q + 2N
.
Integral Formulas with Weight Factors
279
Consequently, q−1 q p q−1 q q−1 dλΩ (ζ) |P1 (z, ζ)| ≤ C(|ρ(z)| 2N +α+n−p+1 ) q Ω
−
= C|ρ(z)| Lemma 4.9
p − n−p+1 − qs q 2N q
.
Let p and N be positive integers with N ≥ 2. Define
Ω = {z ∈ Cn |
p j=1
N +1
|zj |2
+
2p+1
j=p+1
|zj |2 − 1 = ρ(z) < 0}.
Then there exist a constant CN > 0 such that for any holomorphic function f in Ω with f ∈ Lq (Ω) (q ≥ 2), any sufficiently small ε > 0 and θj ∈ [0, 2π], if we define 1
1
z = (ε 2N +1 eiθ1 , · · · , ε 2N +1 eiθp , 0, · · · , 0, 1 − pε), then we obtain − n+2 q −
|f (z)| ≤ CN f Lq (Ω) d(z, ∂Ω)
p 2N q
.
Proof. In Lemma 4.8 we set n = 2p + 1, s = 0. Then we have the desired inequality. Diederich-Mazzilli [DIM1] obtained a counterexample for the Lp (p > 2) extension of holomorphic functions from submanifolds of complex ellipsoids. Theorem 4.10 Given ε > 0, there exist a positive integer p, a complex ellipsoid Ω ⊂ C2p+1 , a submanifold X of C2p+1 which intersects ∂Ω transversally and a bounded holomorphic function f in V = X ∩ Ω such that if g is a holomorphic function in Ω which satisfies g = f in V , then g ∈ L2+ε (Ω). Proof.
Let N be a positive integer. Define V = {z ∈ Ω | z1N + zp+1 = · · · = zpN + z2p = 0}
and f (z) =
z1N −1 · · · zpN −1 (1 − zn )
p(N −1) + q2 2N
.
It follows from the proof of Theorem 4.9 that f is bounded on V . We set 1
1
z = (ε 2N +1 eiθ1 , · · · , ε 2N +1 eiθp , 0, · · · , 0, 1 − pε).
280
Several Complex Variables and Integral Formulas
Using the same method as the proof of Theorem 4.9, for δ > 0 any holomorphic extension g of f to Ω cannot satisfy the inequality |g(z)| ≤ C|ρ(z)|
p(N −1) −1) − p(N +δ 2N 2N +1
.
On the other hand, by Lemma 4.9 if g ∈ Lq (Ω), then we have − p+2 q −
|g(z)| ≤ CgLq (Ω) d(z, ∂Ω)
p 2N q
,
which means that −
p p(N − 1) p(N − 1) p+2 − N ≤ . − q 2 q 2N +1 2N
Then q must satisfy the inequality q≤
2+ 1−
4 1 p + 2N −1 N −1 1 N − 2N
.
Consequently, there is no holomorphic extension g of f which satisfies conditions g ∈ Lq (Ω),
q>
2+ 1−
1 4 p + 2N −1 N −1 1 N − 2N
.
If we choose p and N sufficiently large, then 2+ε>
2+ 1−
4 1 p + 2N −1 N −1 1 N − 2N
,
which implies that there is no holomorphic function g in Ω which satisfies g ∈ L2+ε (Ω) and g|V = f . Mazzilli [MAZ2] investigated Lp extensions of holomorphic functions from submanifolds of complex ellipsoids. Cho [CHO] also obtained a counterexample for the Lp (p > 2) extension of holomorphic functions from submanifolds of some pseudoconvex domain. Tsuji [TSU] gave a counterexample for the bounded extension of holomorphic functions from submanifolds of certain unbounded pseudoconvex domain in C2 .
Integral Formulas with Weight Factors
4.5
281
Bounded Extensions by Means of the Berndtsson Formula
In this section we study the bounded extension of holomorphic functions from complex affine linear hypersurfaces in strictly convex domains. The result has already been proved in 3.3. The aim of the proof is to introduce the method of Diederich-Mazzilli [DIM2] which was used to prove the bounded extension of holomorphic functions from the intersection of a complex affine linear hypersurface with a convex domain of finite type. It is also interesting to compare the method of Henkin-Leiterer (Lemma 3.22) with the method of Diederich-Mazzilli (Theorem 4.12) concerning the integral representation on submanifolds. Let Ω ⊂⊂ Cn be a convex domain with C ∞ boundary. Then there exists a C ∞ function ρ in Cn such that Ω = {z ∈ Cn | ρ(z) < 0}. Let h of Ω. We set X = {z ∈ be a holomorphic function in a neighborhood Ω Ω | h(z) = 0}, V = Ω ∩ X. Since Ω is convex, there exist holomorphic functions gj , j = 1, · · · , n, in Ω × Ω such that h(z) − h(ζ) =
n j=1
Suppose dh = 0 on X. We set
gj (z, ζ)(zj − ζj ).
n
Q1 (z, ζ) =
1 ∂ρ (ζ)dζi . ρ(ζ) i=1 ∂ζi
Then by (4.29) we have the following theorem. Theorem 4.11
Let f be a bounded holomorphic function in V . We define N +n−1 ρ(ζ) ¯ 1 )n−1 Ef (z) := Cn (∂Q f (ζ) < ∂ρ(ζ), z − ζ > +ρ(ζ) V ∧
n ∂h(ζ) ∧ ( j=1 gj (z, ζ)dζj ) ∂h2
dVn−1 (ζ)
for z ∈ Ω, where Cn is a numerical constant depending only on n, and dVn−1 is the Lebesgue measure on V . Then Ef is holomorphic in Ω and satisfies Ef |V = f . Remark 4.2 The integral in the right side of the above equality means to integrate coefficients of forms of degree (n, n) with respect to ζ on V .
Several Complex Variables and Integral Formulas
282
From now on we assume h(z) = zn . For a > 0, we set La = {z | |zn | ≤ a|ρ(z)|, (z1 , · · · , zn−1 , 0) ∈ V }. Since Ω is convex, we have La ⊂ Ω for any sufficiently small a. Let f be a bounded holomorphic function in V . Now we extend f to a C ∞ function in Ω as follows. Let (πγ )γ≥0 be a family of C ∞ functions in R such that πγ ≡ 1 on {x ≤ γ2 }, πγ ≡ 0 on {x ≥ γ}. For z ∈ La , we set f (z1 , · · · , zn ) = f (z1 , · · · , zn−1 , 0). Then f is holomorphic in La . Define ψγ (f )(z) = πγ
|zn |2 ρ(z)2
f (z).
Then for any sufficiently small γ, we have ψγ (f ) ∈ L∞ (Ω) ∩ C ∞ (Ω). Since ¯ γ (f )) = π ′ ∂(ψ γ
|zn |2 ρ(z)2
f (z)∂¯
|zn |2 ρ(z)2
,
we obtain ¯ γ (f )) ≤ ∂(ψ
Cγ , |ρ|
where Cγ is a constant depending on γ such that Cγ → ∞ as γ → 0. Lemma 4.10 Let X = {z | h(z) = 0}, h(z) = zn . Let N be an integer such that N ≥ 2. Suppose s(z, ζ) satisfies the conditions (A) and (B) in Theorem 4.1. Define Q2ε =
n h(ζ)gj (z, ζ) j=1
|h|2
+ε
dζj =
ζ¯n dζn |ζn |2 + ε
zn ζ¯n + ε ¯ 1 n (∂Q ) |ζn |2 + ε ¯ 1 )n−1 ∧ ∂Q ¯ 2 +A2n (< Q1 , z − ζ > +1)−N −n+1 (∂Q ε
P (z, ζ) = A1n (< Q1 , z − ζ > +1)−N −n =: P0 (z, ζ) + P1 (z, ζ)
Integral Formulas with Weight Factors
283
K(z, ζ) = Bn1
n−1 k=0
+Bn2
(< Q1 , z − ζ > +1)−N −k
¯ n−1−k zn ζ¯n + ε ¯ 1 k s ∧ (∂s) ( ∂Q ) ∧ |ζn |2 + ε < s, ζ − z >n−k
n−2 k=0
¯ 1 )k (< Q1 , z − ζ > +1)−N −k (∂Q
¯ n−2−k s ∧ (∂s) ¯ 2 ∧ ∂Q ∧ ε < s, ζ − z >n−k−1 n−1 n−2 K0k (z, ζ) + K1k (z, ζ). =: k=0
k=0
Then we can choose constants Ain , Bni for i = 1, 2 such that ¯ γ (f ))(ζ) ∧ K(z, ζ) ψγ (f )(z) = ψγ (f )(ζ)P (z, ζ) + ∂(ψ Ω
Proof.
(z ∈ Ω).
Ω
In the proof of Theorem 4.7, we set G1 (α) = α−N (N ≥ 2),
G2 (α) = α.
Then we have = An K
α0 +α1 +α2
1 1 , z − ζ > +1)N +α1 (< Q =n−1 ×
P = Bn
α1 +α2
h(ζ)h(z) + ε |h(ζ)|2 + ε
1−α2
¯ α0 ∧ (∂Q ¯ 2 )α2 ¯ 1 )α1 ∧ (∂Q s ∧ (∂s) ε , < s, ζ − z >α0 +1
1 1 , z − ζ > +1)N +α1 (< Q =n
(4.38)
h(ζ)h(z) + ε |h(ζ)|2 + ε
¯ 1 )α1 ∧ (∂Q ¯ 2 )α2 . ×(∂Q ε
×
1−α2
× (4.39)
α2 takes only 0 or 1. By the definition of Q1 , the integral on ∂Ω is equal to 0. Next we assume that Ω ⊂⊂ Cn is a strictly convex domain with C ∞ boundary. Since ρ is strictly convex, it follows from Taylor’s theorem that n ∂ρ 2Re (ζ)(ζj − zj ) ≥ ρ(ζ) − ρ(z) + C|ζ − z|2 ∂ζ j j=1
(z, ζ ∈ Ω).
Several Complex Variables and Integral Formulas
284
Define Φ(z, ζ) =
n ∂ρ (ζ)(ζj − zj ). ∂ζ j j=1
Then we have 2ReΦ(z, ζ) ≥ ρ(ζ) − ρ(z) + C|ζ − z|2
(z, ζ ∈ Ω).
Consequently, 2|Φ(z, ζ) − ρ(ζ)| ≥ |ρ(ζ)| + |ρ(z)| + |ImΦ(z, ζ)| + C|ζ − z|2
(4.40)
for z, ζ ∈ Ω. We take s(z, ζ) as follows: s(z, ζ) = −ρ(z)
n i=1
(ζ¯i − z¯i )dζi − Φ(ζ, z)
n ∂ρ (z)dζi . ∂ζi i=1
For z ∈ ∂Ω\X, ζ ∈ Ω, we have n ∂ρ s(z, ζ) = −Φ(ζ, z) (z)dζi , ∂ζ j j=1
2ReΦ(ζ, z) ≥ −ρ(ζ) + C|ζ − z|2 and s ∧ ∂¯ζ s = 0. Hence for z ∈ ∂Ω\X, if we choose γ sufficiently small, then ψγ (f )(z) = 0, and hence there is no singular point except ζ = z. For z ∈ ∂Ω\X, we obtain ψγ (f )(ζ)P (z, ζ) (4.41) ψγ (f )(z) = Ω
+
Ω
¯ γ (f ))(ζ) ∧ (K n−1 (z, ζ) + K n−2 (z, ζ)). ∂(ψ 0 1
Integral Formulas with Weight Factors
285
¯ γ (f )) = 0 on V , by Lemma 4.5 we have Since ∂(ψ ¯ γ (f ))(ζ) ∧ K n−2 (z, ζ) ∂(ψ lim 1 ε→0 Ω ¯ γ (f )(ζ))πB 2 (< Q1 , z − ζ > +1)−N −n+2 (∂Q ¯ 1 )n−2 ∂(ψ = n V
∧
s dVn−1 = 0. < s, ζ − z >
On the other hand, by Lebesgue’s dominated convergence theorem we have lim ψγ (f ) ∧ P (z, ζ) = 0. γ→0
Ω
By Lemma 4.5 we have lim ψγ (f )P1 (z, ζ) ε→0 Ω ¯ 1 )n−1 dVn−1 = ψγ (f )πA2n (< Q1 , z − ζ > +1)−N −n+1 (∂Q V
= Cn Ef (z).
Thus in (4.41) after letting ε → 0, we let γ → 0. Then we obtain ¯ γ (f ))(ζ)K n−1 (z, ζ), 0 = Cn Ef (z) + lim ∂(ψ 0 γ→0
Ω
where
s ¯ 1 )n−1 ∧ n−1 (z, ζ) = B 1 (< Q1 , z − ζ > +1)−N −n+1 zn (∂Q K . n 0 ζn < s, ζ − z > (4.42) In this setting, we have the following theorem. Theorem 4.12
For z ∈ ∂Ω\X one has 1 ¯ 1 )n−1 Ef (z) = Cn (∂Q zn f (ζ)dζ¯n ∧ 1 , z − ζ > +1)N +n−1 (< Q V s dVn−1 . × < s, ζ − z >
Proof.
n−1 (z, ζ) is expressed by By (4.42), K 0
¯ n−1 (z, ζ) = zn T n−1 (z, ζ) = lim zn ζn T n−1 (z, ζ). K 0 ε→0 |ζn |2 + ε ζn
286
Several Complex Variables and Integral Formulas
It follows from Stokes’ theorem that zn ζ¯n 0= ψ (f )(ζ)T n−1 (z, ζ) 2+ε γ |ζ | n ∂Ω zn ζ¯n ψγ (f )(ζ)∂¯ = ∧ T n−1 (z, ζ) 2+ε |ζ | n Ω zn ζ¯n ¯ + ∂(ψγ (f ))(ζ) ∧ T n−1 (z, ζ) 2 Ω |ζn | + ε zn ζ¯n ¯ n−1 (z, ζ)). + ψγ (f )(ζ)∂(T 2 Ω |ζn | + ε If γ is sufficiently small, then there exists δ > 0 such that ψγ (f ) = 0 in B(z, δ). Since limγ→0 ψγ (f )(ζ) = 0 for ζ ∈ V , by Lebesgue’s dominated convergence theorem we obtain zn ζ¯n ¯ n−1 (z, ζ)) ψγ (f )(ζ)∂(T lim γ→0 Ω |ζn |2 + ε zn ζ¯n ¯ n−1 (z, ζ)) ψγ (f )(ζ)∂(T = lim γ→0 Ω\B(z,δ) |ζn |2 + ε = 0. Consequently, ¯ γ (f ))(ζ)K n−1 (z, ζ) ∂(ψ lim 0 γ→0 Ω = Cn zn f (ζ)dζ¯n ∧ T n−1 (z, ζ)dVn−1 V 1 ¯ 1 )n−1 = Cn zn f (ζ)dζ¯n ∧ (∂Q 1 , z − ζ > +1)N +n−1 (< Q V s dVn−1 . × < s, ζ − z >
The following theorem was proved in 3.3 in more general situation. However, we prove theorem 4.13 using the integral formula in Theorem 4.12. Theorem 4.13 Let Ω ⊂⊂ Cn be a strictly convex domain with C ∞ boundary and let X = {z ∈ Cn | zn = 0}, V = Ω ∩ X. Then every bounded holomorphic function in V can be extended to a bounded holomorphic function in Ω.
Integral Formulas with Weight Factors
287
Proof. Let f be a bounded holomorphic function in V . It is sufficient to show that supz∈∂Ω\X |Ef (z)| ≤ Cf ∞ (see the proof of Theorem 3.15). By Theorem 4.12, if z ∈ ∂Ω\X, then we have ρ(ζ)N +n−1 ¯ 1 )n−1 (∂Q zn f (ζ)dζ¯n ∧ Ef (z) = Cn (−Φ(z, ζ) + ρ(ζ))N +n−1 V n ∂ρ j=1 ∂ζj (z)dζj dVn−1 . × Φ(ζ, z) Taking into account that n−1 ¯ n−1 ¯ n−2 ∧ ∂ρ ∧ ∂ρ ¯ ∂ρ (∂∂ρ) (∂∂ρ) 1 n−1 ¯ = − (n − 1) (∂Q ) = ∂¯ ρ ρn−1 ρn and ¯ ¯ ∂ρ(ζ) ∧ ∂ρ(ζ) ∧ ∂ρ(z) = (∂ρ(ζ) − ∂ρ(z)) ∧ ∂ρ(ζ) ∧ ∂ρ(z) = O(|ζ − z|), it is sufficient to estimate the following two integrals: |ζ − z| I1 = dVn−1 (ζ) (|ρ(ζ)| + |Φ(z, ζ)|)n |Φ(ζ, z)| V and I2 =
V
1 dVn−1 (ζ). (|ρ(ζ)| + |Φ(z, ζ)|)n−1 |Φ(ζ, z)|
By Lemma 3.43 we have Φ(ζ, z) = Φ(z, ζ) − ρ(ζ) + O(|ζ − z|3 ). It follows from (4.40) that dVn−1 I1 ≤ C . 2 3 2n−5 V (|ρ(ζ)| + |ImΦ(z, ζ)| + |ζ − z| ) |ζ − z| We choose a coordinate system t1 , t2 , · · · , t2n−2 such that t1 = ρ(ζ) − ρ(z), t2 = Im Φ(z, ζ) and |t| ≈ |ζ − z|. We set t′ = (t3 , · · · , t2n−2 ). Then we have dt I1 ≤ C 2 ′ 2 3 ′ 2n−5 |t|≤R (|t1 | + |t2 | + |zn | + |t | ) |t | dt3 · · · dt2n−2 ≤C 2 ′ 2 ′ 2n−5 |t′ | 0, q ≥ 1 and z = x + iy, prove that |z + w|j dxdy O(A1−q ) (q > 1) = j 2 q O(log A) (q = 1). |z| 0 such that for t, τ ∈ R t2p − τ 2p − 2pτ 2p−1 (t − τ ) ≥ δ{τ 2p−2 (t − τ )2 + (t − τ )2p } 4.5
Let m be a positive integer. For σ > 0, define Γσ = {z = x+iy | |y| < σ|x|}. Prove that there exist σ > 0 and ε > 0 such that Re(z 2m ) ≥ ε|z|2m on Γσ .
4.6
Prove the following: (a) For q ≥ 1, l = 0 or 1, j = l , l + 1, · · · , and A positive close to 0 |t + x|j−l |x|l dxdy O(A1−q ) (q > 1) , = j 2 2 q O(log A) (q = 1) |z| 1) = , j 2 j+2 )q O(log A) (q = 1) |z| 0, define Pj (z, ζ) =
∂ρ 2m −2 2n −2 (ζ) − γ[(ηj j − ξj j )(zj − ζj ) + (zj − ζj )2mj −1 ] ∂zj
Several Complex Variables and Integral Formulas
290
and Φ(z, ζ) =
N j=1
Pj (z, ζ)(zj − ζj ).
Prove the following (1) and (2). (1) If we choose γ > 0 sufficiently small, then there exists ε > 0 such that 2ReΦ(z, ζ) ≤ ρ(z) − ε
N
{(ξk2nk −2 + ηk2mk −2 )|zk − ζk |2
k=1
+|zk − ζk |2mk } for (z, ζ) ∈ Ω × ∂Ω. (2) Let q = maxj min{2nj , 2mj }. Then there exists a constant C > 0 such that for every bounded, ∂¯ closed (0, 1) form f on Ω, there ¯ = f (in exists a 1/q-H¨older continuous function u in Ω such that ∂u the sense of distributions) and u1/q ≤ Cf 0,Ω . (See DiederichFornaess-Wiegerinck [DIK]).
Chapter 5
The Classical Theory in Several Complex Variables
In this chapter we first prove the Poincar´e theorem, and then we investigate the Weierstrass preparation theorem, the properties of the coherent analytic sheaf and the Cousin problem. Some of theorems in Chapter 5 were used to prove the theorems in the previous chapters.
5.1
The Poincar´ e Theorem
In this section we study the Poincar´e theorem which says that there is no biholomorphic mapping from a ball to a polydisc in Cn (n ≥ 2). Here we give the proof due to Krantz [KR3]. Definition 5.1 Let B be the unit disc in the complex plane. Let Ω be a domain in C or C2 . For P ∈ Ω, define (B, Ω)P := {f : Ω → B | f is holomorphic, f (P ) = 0} and (Ω, B)P := {f : B → Ω | f is holomorphic, f (0) = P }. Moreover, we define for Ω ⊂ C2 and f ∈ (B, Ω)P ∂f ∂f (P ), (P ) . JacC f (P ) = ∂z1 ∂z2 Definition 5.2 (a) For P ∈ Ω ⊂ C2 , ξ ∈ C2 , we define the Carath´eodory metric FCΩ (P, ξ) of ξ at P by FCΩ (P, ξ) = sup{|JacC f (P )ξ| | f ∈ (B, Ω)P }. 291
Several Complex Variables and Integral Formulas
292
(b) For P ∈ Ω ⊂ C, ξ ∈ C, we define the Carath´eodory metric FCΩ (P, ξ) of ξ at P by FCΩ (P, ξ) = sup{|f ′ (P )ξ| | f ∈ (B, Ω)P }. Definition 5.3 (a) Let Ω ⊂ C2 be a domain. For P ∈ Ω, ξ ∈ C2 , we Ω (P, ξ) of ξ at P by define the Kobayashi metric FK Ω (P, ξ) FK |ξ| ′ = inf | g ∈ (Ω, B) , there exists λ such that g (0) = λ . P |g ′ (0)|
(b) Let Ω ⊂ C be a domain. For P ∈ Ω, ξ ∈ C, we define the Kobayashi Ω metric FK (P, ξ) of ξ at P by |ξ| Ω | g ∈ (Ω, B)P . FK (P, ξ) = inf |g ′ (0)| Theorem 5.1 Let Ω1 ⊂ C2 and Ω2 ⊂ Cn , 1 ≤ n ≤ 2, be domains and let f : Ω1 → Ω2 be a holomorphic mapping. For P ∈ Ω1 and ξ ∈ C2 , define f∗ (P )ξ = JacC f (P )ξ. Then we have FCΩ1 (P, ξ) ≥ FCΩ2 (f (P ), f∗ (P )ξ)
(5.1)
Ω1 Ω2 FK (P, ξ) ≥ FK (f (P ), f∗ (P )ξ).
(5.2)
and
Proof. obtain
Let n = 2. Let ϕ ∈ (B, Ω2 )f (P ) . Then ϕ ◦ f ∈ (B, Ω1 )P . We FCΩ1 (P, ξ) ≥ JacC (ϕ ◦ f )(P )ξ ∂(ϕ ◦ f ) ∂(ϕ ◦ f ) ξ1 = (P ), (P ) ξ2 ∂z1 ∂z2 ∂ϕ ∂ϕ = (f (P )), (f (P )) f∗ (P )ξ ∂w1 ∂w2 = |JacC ϕ(f (P ))f∗ (P )ξ| .
Since ϕ ∈ (B, Ω2 )f (P ) is arbitrary, we have (5.1). When n = 1 we can prove (5.1) similarly. Next we prove (5.2). Let n = 2. Let g ∈ (Ω1 , B)P
The Classical Theory in Several Complex Variables
293
and g ′ (0) = λξ. Then f ◦ g ∈ (Ω2 , B)f (P ) . We set g = (g1 , g2 ). Then we have (f ◦ g)′ (0) ∂f1 ∂f1 ∂f2 ∂f2 = (P )g1′ (0) + (P )g2′ (0), (P )g1′ (0) + (P )g2′ (0) ∂z1 ∂z2 ∂z1 ∂z2 = λf∗ (P )ξ. Consequently, Ω2 (f (P ), f∗ (P )ξ) FK
|f∗ (P )ξ| | h ∈ (Ω2 , B)f (P ) , h′ (0) = µf∗ (P )ξ = |h′ (0)| |f∗ (P )ξ| 1 |ξ| ≤ = = ′ . ′ |(f ◦ g) (0)| |λ| |g (0)|
Since g is arbitrary, we have (5.2). When n = 1, we can prove (5.2) in the same way. Corollary 5.1 Let Ω1 and Ω2 be domains in C2 and let f : Ω1 → Ω2 be a biholomorphic mapping. Then FCΩ1 (P, ξ) = FCΩ2 (f (P ), f∗ (P )ξ) and Ω1 Ω2 FK (P, ξ) = FK (f (P ), f∗ (P )ξ)
for P ∈ Ω1 and ξ ∈ C2 . Proof. For f −1 : Ω2 → Ω1 , f (P ) ∈ Ω2 and η ∈ C2 , we apply Theorem 5.1. Then we have FCΩ2 (f (P ), η) ≥ FCΩ1 (f −1 (f (P )), (f −1 )∗ (f (P ))η) and Ω2 Ω1 (f (P ), η) ≥ FK (f −1 (f (P )), (f −1 )∗ (f (P ))η). FK
If we set η = f∗ (P )ξ, then we have FCΩ2 (f (P ), f∗ (P )ξ) ≥ FCΩ1 (P, ξ) and Ω2 Ω1 (f (P ), f∗ (P )ξ) ≥ FK (P, ξ). FK
Together with (5.1) and (5.2) we have the desired equalities.
Several Complex Variables and Integral Formulas
294
Definition 5.4 Let Ω ⊂ C2 be a domain. We define the length of a C 1 curve γ : [a, b] → Ω with respect to the Carath´eodory metric and the Kobayashi metric by b FCΩ (γ(t), γ ′ (t))dt lC (γ) = a
and lK (γ) =
b
a
Ω FK (γ(t), γ ′ (t))dt,
respectively. Let f : Ω1 → Ω2 be a holomorphic mapping. Then
Corollary 5.2
lC (f ◦ γ) ≤ lC (γ),
lK (f ◦ γ) ≤ lK (γ)
for every C 1 curve γ : [a, b] → Ω1 . Proof.
Let f = (f1 , f2 ). Then we have (f ◦ γ)′ (t) = ((f1 ◦ γ)′ (t), (f2 ◦ γ)′ (t)) = f∗ (γ(t))γ ′ (t).
By Theorem 5.1 we obtain lK (f ◦ γ) = =
b
a
b
a
≤
a
b
Ω2 FK (f ◦ γ(t), (f ◦ γ)′ (t))dt Ω2 FK (f (γ(t)), f∗ (γ(t))γ ′ (t)dt Ω1 FK (γ(t), γ ′ (t))dt = lK (γ).
We can prove lC (f ◦ γ) ≤ lC (γ) similarly. Definition 5.5
Let Ω ⊂ C2 be a domain. For P ∈ Ω, define 2 Ω iC P (Ω) = {ξ ∈ C | FC (P, ξ) < 1}
and 2 Ω iK P (Ω) = {η ∈ C | FK (P, η) < 1}.
Theorem 5.2 Let Ω1 and Ω2 be domains in C2 and let f : Ω1 → Ω2 be a biholomorphic mapping. For P ∈ Ω1 , we set Q = f (P ). Then linear mappings C JacC f (P ) : iC P (Ω1 ) → iQ (Ω2 )
The Classical Theory in Several Complex Variables
295
and K JacC f (P ) : iK P (Ω1 ) → iQ (Ω2 )
are bijective. Proof. We set g(ξ) = JacC f (P )ξ. It follows from Corollary 5.1 that for ξ ∈ iC P (Ω1 ) FCΩ2 (Q, g(ξ)) = FCΩ2 (f (P ), f∗ (P )ξ) = FCΩ1 (P, ξ) < 1, C which implies that g(ξ) ∈ iC Q (Ω2 ). Clearly g is linear. For η ∈ iQ (Ω2 ) we −1 set h(η) = JacC f (Q)η. Then we have
FCΩ1 (P, h(η)) = FCΩ1 (f −1 (Q), JacC f −1 (Q)η) = FCΩ2 (Q, η) < 1, −1 which means that h(η) ∈ iC (w) = w with P (Ω1 ). Differentiating f ◦ f −1 respect to w1 and w2 , we have JacC f (P )JacC f (Q) = E (E is the unit matrix). Similarly, we have JacC f −1 (Q)JacC f (P ) = E. Consequently, we C have g ◦ h(η) = η, h ◦ g(ξ) = ξ. Hence g : iC P (Ω1 ) → iQ (Ω2 ) is bijective. K K Similarly, JacC f (P ) : iP (Ω1 ) → iQ (Ω2 ) is bijective.
Lemma 5.1
Let a and b be complex numbers such that |az1 + bz2 | ≤ 1
for any complex numbers z1 , z2 with |z1 |2 + |z2 |2 = 1. Then |a|2 + |b|2 ≤ 1. Proof. Let a = r1 eiθ1 , b = r2 eiθ2 . For z1 = t1 e−iθ1 , z2 = t2 e−iθ2 with t21 + t22 = 1, we set t1 = cos θ, t2 = sin θ. Then we have 1 ≥ |az1 + bz2 | = t1 r1 + t2 r2 = r1 cos θ + r2 sin θ r1 r2 2 2 = (r1 + r2 ) 2 cos θ + 2 sin θ r1 + r22 r1 + r22 = (r12 + r22 ) sin(θ + α).
We can choose t1 and t2 in such a way that θ + α = r12 + r22 ≤ 1. Theorem 5.3 B(0, 1).
π 2,
which means that
Let B(0, 1) = {z ∈ C2 | |z| < 1}. Then iK 0 (B(0, 1)) =
Several Complex Variables and Integral Formulas
296
Proof. Let B = {z ∈ C | |z| < 1} and let ϕ ∈ (B(0, 1), B)0 . For η ∈ C2 with |η| = 1, define h(ζ) = ϕ(ζ) · η
(ζ ∈ B).
Then h : B → B satisfies ϕ(0) = 0. It follows from the Schwarz lemma that |h′ (0)| ≤ 1. Since η is arbitrary so far as |η| = 1, it follows from Lemma 5.1 that |ϕ′ (0)| ≤ 1. Consequently, |ξ| B(0,1) | ϕ ∈ (B(0, 1), B)0 ≥ |ξ|. FK (0, ξ) = inf |ϕ′ (0)| for every ξ ∈ C2 . On the other hand, for ξ = 0 we set ϕ0 (ζ) =
ζ ξ |ξ|
(ζ ∈ B).
Then ϕ0 ∈ (B(0, 1), B)0 , which implies that B(0,1)
FK B(0,1)
Hence we have FK
(0, ξ) ≤
|ξ|
|ϕ′0 (0)|
= |ξ|.
(0, ξ) = |ξ|, and hence iK 0 (B(0, 1)) = B(0, 1).
Define P (0, 1) = {z ∈ C2 | |z1 | < 1, |z2 | < 1}. Then
Theorem 5.4
iK 0 (P (0, 1)) = P (0, 1). Proof.
Define mappings π1 : P (0, 1) → B and π2 : P (0, 1) → B by π1 (z1 , z2 ) = z1 ,
π2 (z1 , z2 ) = z2 .
For η = (η1 , η2 ) ∈ C2 , it follows from Theorem 5.1 that P (0,1)
FK
B B (0, η) ≥ FK (π1 (0), (π1 )∗ η) = FK (0, η1 ).
By the Schwarz lemma, for a holomorphic mapping ϕ : B → B with ϕ(0) = 0, we have |ϕ′ (0)| ≤ 1. Moreover, if we define ϕ0 : B → B by ϕ0 (ζ) = ζ, then we have ϕ′ (0) = 1. Hence we have |η1 | B FK (0, η1 ) = | ϕ ∈ (B, B)0 = |η1 |. |ϕ′ (0)|
The Classical Theory in Several Complex Variables
297
Consequently, P (0,1)
FK
(0, η) ≥ |η1 |
and P (0,1)
FK
(0, η) ≥ |η2 |.
Therefore we obtain P (0,1)
FK
(0, η) ≥ max{|η1 |, |η2 |},
2 which means that iK 0 (P (0, 1)) ⊂ P (0, 1). Next, for 0 = η ∈ C , we set ζη1 ζη2 , ψ(ζ) = . max{|η1 |, |η2 |} max{|η1 |, |η2 |}
Then ψ ∈ (P (0, 1), B)0 and we have ψ ′ (0) = µη (µ > 0). Consequently, P (0,1)
FK
(0, η) ≤
|η| = max{|η1 |, |η2 |}, |ψ ′ (0)|
which means that iK 0 (P (0, 1)) ⊃ P (0, 1).
Theorem 5.5 (Poincar´ e theorem) There is no biholomorphic mapping from the unit polydisc P (0, 1) in C2 onto the unit ball B(0, 1) in C2 . Proof. Suppose there is a biholomorphic mapping Φ : P (0, 1) → B(0, 1). We set Φ−1 (0) = α. Then α ∈ P (0, 1). Let α = (α1 , α2 ). Then α1 ∈ B, α2 ∈ B. For z ∈ B, we set ϕ1 (z) = and
z − α1 , 1−α ¯1 z
ϕ2 (z) =
z − α2 1−α ¯2z
ϕ(ζ) = (ϕ1 (ζ1 ), ϕ2 (ζ2 )). Then ϕ : P (0, 1) → P (0, 1) is holomorphic and bijective. Further we have ϕ(α) = 0. Define g = Φ ◦ ϕ−1 . Then g : P (0, 1) → B(0, 1) is biholomorphic and g(0) = 0. Next we show that g does not exist. By Theorem 5.2, K JacC g(0) is a bijective linear mapping from iK 0 (P (0, 1)) onto i0 (P (0, 1)). By Theorem 5.3 and Theorem 5.4, JacC g(0) is a bijective linear mapping from P (0, 1) onto B(0, 1). We set h = JacC g(0). Since h : P (0, 1) → B(0, 1) is biholomorphic, h maps ∂P (0, 1) to ∂B(0, 1). Therefore a segment A = {(t, 1) | 0 ≤ t ≤ 1} ⊂ ∂P (0, 1) is mapped by h to ∂B(0, 1). Since h is
298
Several Complex Variables and Integral Formulas
linear, h(A) is also a segment. Since B is strictly convex, ∂B(0, 1) cannot contain a segment (see Lemma 3.12), which is a contradiction. Hence a biholomorphic mapping Φ does not exist. 5.2
The Weierstrass Preparation Theorem
We prove the Weierstrass preparation theorem using the Cauchy integral formula. Further we prove the implicit function theorem for holomorphic functions. Definition 5.6 Let f be a holomorphic function in a neighborhood of a ∈ Cn , and let f (a) = 0. We set a = (a′ , an ). We say that f is regular of order k in zn at the point a if f (a′ , zn ), considered as a holomorphic function of the single variable zn , has a zero of order k at the point zn = an . Equivalently, the condition can be stated as follows: g(an ) = g ′ (an ) = · · · = g (k−1) (an ) = 0,
g (k) (an ) 6= 0,
where g(zn ) = f (a′ , zn ). Lemma 5.2 Let f be a holomorphic function in B(a, ε) and let f (a) = 0, f (z) 6≡ 0. Then after a suitable complex linear change of coordinates in Cn , the function will be regular of order k, k ≥ 1, in zn at the point a. Proof. There exists p ∈ B(a, ε), p 6= a, such that f (p) 6= 0. There exist constants bij such that the linear change of coordinates zi = (pi − ai )(ζn − an ) +
n−1 X
bij (ζj − aj ) + ai
(i = 1, · · · , n)
j=1
is nonsingular. We set g(ζ) = f (z(ζ)). Then g(a1 , · · · , an−1 , 1 + an ) = f (p1 , · · · , pn ) 6= 0, g(a1 , · · · , an−1 , an ) = f (a1 , · · · , an ) = 0. we set h(ζn ) = g(a1 , · · · , an−1 , ζn ). Then by the identity theorem, h(ζn ) has a zero of order k at the point ζn = an for some positive integer k. Hence g is regular of order k in ζn at a. Lemma 5.3 Let f be a holomorphic function in a neighborhood of 0 and let f (0) = 0. Let f be regular of order k in zn at 0, k ≥ 1. Then for each sufficiently small δn > 0 there is δ ′ = (δ1 , · · · , δn−1 ), such that for
The Classical Theory in Several Complex Variables
299
each fixed z ′ ∈ P (0′ , δ ′ ), the equation f (z ′ , zn ) = 0 has precisely k solutions (counted with multiplicities) in the disc {|zn | < δn }. Proof. We set g(zn ) = f (0′ , zn ). By the assumption g(zn ) has a zero of order k at the point 0. Since the zero set of any non-constant holomorphic function in one variable is discrete, there exists δn > 0 such that g(zn ) is holomorphic and nowhere vanishing in {0 < |zn | ≤ δn }. We set m = min|zn |=δn |g(zn )|. If we choose δ ′ = (δ1 , · · · , δn−1 ) sufficiently small, then for z ′ ∈ P (0′ , δ ′ ), |zn | ≤ δn , using the uniform continuity we have |f (0′ , zn ) − f (z ′ , zn )| < m. Hence for |zn | = δn we obtain |g(zn )| ≥ m > |g(zn ) − f (z ′ , zn )|. By the Rouch´e theorem, the number of zeros of g(zn ) in {|zn | < δn } counting multiplicities equals the number of zeros of f (z ′ , zn ) in {|zn | < δn } counting multiplicities. Since g(zn ) has a zero of order k in 0 and does not vanish except 0, f (z ′ , zn ) has k zeros in {|zn | < δn }. Let f be a holomorphic function in a neighborhood of 0 ∈ Cn and let f (0) = 0. Suppose f is regular of order k in zn (k ≥ 1) at 0. By Lemma 5.3, there exists δ = (δ1 , · · · , δn ) such that for any z ′ ∈ P (0′ , δ ′ ) f (z ′ , ·) has k zeros ϕ1 (z ′ ), · · · , ϕk (z ′ ) in {|zn | < δn } counting multiplicities. We set ω(z ′ , zn ) = (zn − ϕ1 (z ′ )) · · · (zn − ϕk (z ′ )) = znk + ak−1 (z ′ )znk−1 + · · · + a0 (z ′ ). Then the number of zeros of f (z ′ , ·) in {|zn | < δn } counting multiplicities equals the number of zeros of ω(z ′ , ·) in {|zn | < δn } counting multiplicities. Hence there exists a nonvanishing function u in P (0, δ) such that f = ωu. The Weierstrass preparation theorem says that we can choose ω and u to be holomorphic. Moreover, since |ϕj (z ′ )| < δn , we have ω(z ′ , zn ) = 0 for z ′ ∈ P (0′ , δ ′ ), |zn | = δn . Definition 5.7
A function
ω(z ′ .zn ) = znk + ak−1 (z ′ )znk−1 + · · · + a0 (z ′ ) is called a Weierstrass polynomial at 0 if aj (z ′ ) for j = 0, · · · , k −1 are holomorphic in a neighborhood of 0′ and satisfies aj (0′ ) = 0 for j = 0, · · · , k −1. Theorem 5.6 (Weierstrass preparation theorem) Let f be a holomorphic function in a neighborhood of 0 ∈ Cn and let f (0) = 0. Suppose f
300
Several Complex Variables and Integral Formulas
is regular of order k, k ≥ 1, in zn at 0. Then there exists δ = (δ1 , · · · , δn ) such that for z ∈ P (0, δ) we have a unique factorization f (z) = ω(z)u(z), where ω(z) = znk + ak−1 (z ′ )znk−1 + · · · + a0 (z ′ ) is a Weierstrass polynomial at 0 and u is a nowhere vanishing holomorphic function in P (0, δ). Proof. The uniqueness is obvious. By Lemma 5.3 there exist δ = (δ ′ , δn ) such that for any fixed z ′ ∈ P (0, δ ′ ), f (z ′ , ·) has k zeros ϕ1 (z ′ ), · · · , ϕk (z ′ ) in {|zn | < δn }. By the argument principle, we have for z ′ ∈ P (0, δ ′ ) Sm (z ′ ) ≡
k j=1
′ ϕm j (z ) =
1 2πi
|ζ|=δn
′ ζ m ∂f ∂ζ (z , ζ)
f (z ′ , ζ)
dζ.
Hence Sm (z ′ ) is holomorphic as a function in z ′ . Moreover, a0 , · · · , ak are holomorphic since they are polynomials of S0 , · · · , Sk−1 . Since f (0′ , zn ) = znk g(zn ) and g(0) = 0, we have ϕ1 (0′ ) = · · · = ϕk (0′ ) = 0. Hence a0 (0′ ) = · · · = ak−1 (0′ ) = 0, which means that ω(z) = znk + ak−1 (z ′ )znk−1 + · · · + a0 (z ′ ) is a Weierstrass polynomial at 0. We set u = f /ω. Since u(z ′ , ·) is holomorphic in {|zn | ≤ δn } for fixed z ′ , we obtain (f /ω)(z ′ , ζ) 1 ′ u(z , zn ) = dζ. 2πi |ζ|=δn ζ − zn Since ω(z ′ , zn ) = 0 for |zn | = δn , u is holomorphic in P (0, δ).
Theorem 5.7 (Weierstrass division theorem) Let ω be a Weierstrass polynomial of degree k at 0, and let f be holomorphic in a neighborhood of 0 ∈ Cn . Then there is a unique factorization in some sufficiently small neighborhood of 0 f = ωq + r, where q and r are holomorphic in a neighborhood of 0 and r is a polynomial in zn of degree less than k with coefficients that are holomorphic functions of z1 , · · · , zn−1 . Proof. As we mentioned in the remark before Theorem 5.6, there exists δ = (δ1 , · · · , δn ) = (δ ′ , δn ) such that if z ′ ∈ P (0′ , δ ′ ), |zn | = δn , then ω(z ′ , zn ) = 0. We set f (z ′ , ζ) 1 q(z ′ , zn ) = dζ. ′ 2πi |ζ|=δn ω(z , ζ)(ζ − zn )
The Classical Theory in Several Complex Variables
301
Then q is holomorphic in P (0, δ). Define r = f − qω. Then r is holomorphic in P (0, δ). Consequently, we have 1 dζ f (z ′ , ζ) ′ ′ r(z) = f (z , ζ) − ω(z , zn ) ′ 2πi |ζ|=δn ω(z , ζ) ζ − zn ′ ′ f (z , ζ) ω(z , ζ) − ω(z ′ , zn ) 1 dζ. = 2πi |ζ|=δn ω(z ′ , ζ) ζ − zn Taking into account that k−1 (ζ k − znk ) + j=0 aj (z ′ )(ζ j − znj ) ω(z ′ , ζ) − ω(z ′ , zn ) = , ζ − zn ζ − zn r is a polynomial in zn of degree less than k. Next we show the uniqueness. Suppose we have two factorizations q1 ω + r1 = q2 ω + r2 = f. In the equation r1 − r2 = (q2 − q1 )ω, the left side is a polynomial in zn of degree less than k and the right side has k zeros in zn , which means that r1 ≡ r2 . Hence we have q1 = q2 . Theorem 5.8 Let f be a holomorphic function in a polydisc P (w, r) ⊂ Cn with the following properties: (a) f (w) = 0. ∂f (w) = 1. (b) ∂z n Let w = (w′ , wn ). Then there exist δ = (δ1 , · · · , δn ) = (δ ′ , δn ) and a holomorphic function ϕ(z1 , · · · , zn−1 ) in a polydisc P (w′ , δ ′ ) such that in P (w, δ), f (z1 , · · · , zn ) = 0 is equivalent to zn = ϕ(z1 , · · · , zn−1 ). Proof.
We set f = u + iv, x = (x1 , · · · , x2n ),
zj = xj + ixn+j
(j = 1, · · · , n),
x′ = (x1 , · · · , xn−1 , xn+1 , · · · , x2n ).
Several Complex Variables and Integral Formulas
302
Then by the Cauchy-Riemann equations we have 2 ∂f ∂(u, v) ∂zn (w) = ∂(xn , x2n ) (w) .
By the implicit function theorem in real variables, there exist C ∞ functions g(x′ ), h(x′ ) in a neighborhood of w′ such that u(x) = 0, v(x) = 0 are equivalent to xn = g(x′ ), x2n = h(x′ ). We set ϕ(z ′ ) = g(x′ ) + ih(x′ ). Then in a neighborhood of w, f (z) = 0 is equivalent to zn = ϕ(z ′ ), which implies that f (z1 , · · · , zn−1 , ϕ(z ′ )) = 0. By the condition (b) and Lemma 5.3, there exists δ = (δ1 , · · · , δn ) such that for z ′ = (z1 , · · · , zn−1 ) ∈ P (w′ , δ ′ ), f (z ′ , ·) has only zero ϕ(z ′ ) of order 1 in |wn − zn | < δn . By the argument principle, we have 1 ϕ(z ) = 2πi ′
Z
′
,ζn ) ζn ∂f (z ∂ζn
|ζn −wn |=δn
f (z ′ , ζn )
dζn .
Therefore ϕ(z ′ ) is holomorphic in P (w′ , δ ′ ).
Next we prove the implicit function theorem for holomorphic functions. Theorem 5.9 (Implicit function theorem) Let 1 ≤ k ≤ n. Let fk+1 , · · · , fn be holomorphic functions in a polydisc P (w, r) ⊂ Cn satisfying the following properties: (a) fj (w) = 0 ∂f (b) ∂zji (w) = δij
(j = k + 1, · · · , n). (i, j = k + 1, · · · , n).
Let w = (w′′ , wk+1 , · · · , wn ). Then there exist δ = (δ ′′ , δk+1 , · · · , δn ) and holomorphic functions ϕj (z1 , · · · , zk ) for j = k + 1, · · · , n in a polydisc P (w′′ , δ ′′ ) such that in a neighborhood of w, equations fj (z1 , · · · , zn ) = 0 for j = k + 1, · · · , n are equivalent to equations zj = ϕj (z1 , · · · , zk ) for j = k + 1, · · · , n. Proof. We prove Theorem 5.9 by induction on m = n − k. In case m = 1, Theorem 5.9 follows from Theorem 5.8. Assume that Theorem 5.9 has already been proved for m − 1. Suppose fk+1 , · · · , fn are holomorphic in P (w, r) and satisfy conditions (a) and (b). We set z = (z ′ , zn ), w = (w′ , wn ). We apply Theorem 5.8 to fn (z). Then there exist η = (η ′ , ηn ) and a holomorphic function ϕ(z ′ ) in P (w′ , η ′ ) such that in P (w, η), an equation f (z ′ , zn ) = 0 is equivalent to zn = ϕ(z ′ ). We set gj (z ′ ) = fj (z ′ , ϕ(z ′ ))
(j = k + 1, · · · , n − 1).
The Classical Theory in Several Complex Variables
303
Then fj (z) = 0, j = k + 1, · · · , n, are equivalent to gj (z ′ ) = 0
(j = k + 1, · · · , n − 1), zn = ϕ(z ′ ).
Consequently, we have gj (w′ ) = 0, j = k + 1, · · · , n − 1. For i = k + 1, · · · , n − 1 we have ∂gj ′ ∂fj ∂fj ∂ϕ ′ (w ) = (w) + (w) (w ) = δij . ∂zi ∂zi ∂zn ∂zi By the inductive hypothesis, there are δ ′ = (δ ′′ , δk+1 , · · · , δn−1 ) and holomorphic functions ϕk+1 (z ′′ ), · · · , ϕn−1 (z ′′ ) in P (w′′ , δ ′′ ) such that equations gj (z ′ ) = 0, j = k + 1, · · · , n − 1, are equivalent to equations zj = ϕj (z ′′ ), j = k + 1, · · · , n − 1 in P (w′ , δ ′ ). We set ϕn (z ′′ ) = ϕ(z ′′ , ϕk+1 (z ′′ ), · · · , ϕn−1 (z ′′ )). Then equations fj (z) = 0, j = k + 1, · · · , n, are equivalent to equations zj = ϕj (z ′′ ), j = k + 1, · · · , n. Definition 5.8 Let Ω be a domain in Cn , and let F = (f1 , · · · , fm ) : Ω → Cm be a holomorphic mapping. Define ∂fi F ′ (z) = (z) . ∂zj F ′ (z) is called the Jacobian matrix of F at z ∈ Ω. We say that F is nonsingular at z if the rank of F ′ (z) equals min(m, n). Theorem 5.10 Let Ω ⊂ Cn be a domain with 0 ∈ Ω. Let n ≥ m. Suppose F : Ω → Cm is a nonsingular holomorphic mapping and F (0) = 0. Pn Then there exist a linear change of variables wi = j=1 aij zj for i = 1, . . . , n, δ = (δ ′ , δn−m+1 , · · · , δn ) where δ ′ = (δ1 , · · · , δn−m ) and holomorphic functions ϕj (w1 , · · · , wn−m ) for j = n − m + 1, · · · , n in P (0.δ ′ ) such that in P (0, δ), equations F (w1 , · · · , wn ) = 0 are equivalent to equations wj = ϕj (w1 , · · · , wn−m ), j = n − m + 1, · · · , n. Proof. Let F = (fn−m+1 , · · · , fn ). By the hypothesis there exist an (m × m) matrix B and an (n × n) matrix A such that BF ′ (0)A−1 = (0, I),
304
Several Complex Variables and Integral Formulas
where I is an (m × m) unit matrix. We set B = (bij )
A−1 = (a′ij )
A = (aij ),
gi =
n
(n − m + 1 ≤ i, j ≤ n),
bij fj
j=n−m+1
wi =
n
(1 ≤ i, j ≤ n),
(i = n − m + 1, · · · , n),
aij zj
j=1
(i = 1, · · · , n).
Further we set G = (gn−m+1 , · · · , gn ). Then F (z) = 0 is equivalent to G(z) = 0. Taking into account that ∂gi = ∂wj
n
k=n−m+1
bik
∂fk = ∂wj
n
k=n−m+1
bik
n ∂fk s=1
∂zs
a′sj ,
we obtain ∂gi (0) = δji . ∂wj Theorem 5.10 follows from Theorem 5.9.
Corollary 5.3 (Inverse mapping theorem) Let F be a nonsingular holomorphic mapping from a neighborhood of z ∈ Cn into Cn and let F (z) = w. Then there exist a neighborhood U (U ⊂ W ) of z and a neighborhood V of w such that F : U → V has the holomorphic inverse mapping F −1 : V → U . Proof. We assume that z = w = 0. We denote by J(0) the Jacobian matrix of F at 0. Without loss of generality we may assume that J(0) is the unit matrix. We set H(z, w) = w − F (z). By Theorem 5.10, there exist ε > 0, a polydisc P (0, ε) in Cn and a holomorphic mapping G in P (0, ε) such that the equation H(z, w) = 0 is equivalent to the equation z = G(w). Therefore we have w = F ◦ G(w), z = G ◦ F (z), which implies that G = F −1 . Theorem 5.11 Let Ω ⊂ Cn be an open set. If a holomorphic mapping F : Ω → Cn is injective, then detF ′ (z) = 0 for z ∈ Ω.
The Classical Theory in Several Complex Variables
305
Proof. We prove Theorem 5.11 by induction on n. When n = 1, Theorem 5.11 is true. Assume that Theorem 5.11 has already been proved for n − 1. Now we prove the following lemma: Lemma 5.4 Let Ω ⊂ Cn be an open set. Suppose a holomorphic mapping F : Ω → Cn is injective. If F ′ (a) = 0 for a ∈ Ω, then detF ′ (a) = 0. Proof of Lemma 5.4 Without loss of generality, we may assume that n F = (f1 , · · · , fn ), ∂f ∂zn (a) = 0. We set w(z) = (z1 , · · · , zn−1 , fn (z)). Then k det ∂w (a) = 0. By Corollary 5.3, w−1 is a holomorphic mapping in a ∂zj neighborhood of a. We set F = F ◦ w−1 . Then we have F (w) = (g1 (w), · · · , gn−1 (w), wn ),
where g1 , · · · , gn−1 are holomorphic at b = w(a). Set w′ = (w1 , · · · , wn−1 ) and G(w′ ) = (g1 (w′ , bn ), · · · , gn−1 (w′ , bn )). Then G is injective in a neighborhood of b′ = (b1 , · · · , bn−1 ). By the inductive hypothesis, we have detG′ (b′ ) = 0, which implies that detF′ (b) = 0. Hence we have detF ′ (a) = 0, which completes the proof of Lemma 5.4. We continue the proof of Theorem 5.11. We set h = detF ′ ∈ O(Ω). Suppose Z(h) = {z ∈ Ω | h(z) = 0} = φ. Then Z(h) contains a n − 1 dimensional submanifold M . By Lemma 5.4, we have F ′ (z) = 0 for z ∈ Z(h), and hence F ′ (z) = 0 for z ∈ M . Consequently, F is locally constant in M . Since dimC M = n − 1 > 0, this contradicts that F is injective. Corollary 5.4 Let Ω ⊂ Cn be an open set. If a holomorphic mapping F : Ω → Cn is injective, then F (Ω) is an open set and F −1 : F (Ω) → Ω is holomorphic.
Proof. By Theorem 5.11, we have detF ′ (z) = 0 for z ∈ Ω. Hence F (Ω) is open. By Corollary 5.3, F −1 : F (Ω) → Ω is holomorphic.
Theorem 5.12 Let Ω be a pseudoconvex domain in Cn and let m be a positive integer with m ≤ n. Suppose that f1 , · · · , fm are holomorphic functions in Ω and that F = (f1 , · · · , fm ) is nonsingular in Ω. We set M = {z ∈ Ω | f1 (z) = · · · = fm (z) = 0}. Let a ∈ Ω. If f is holomorphic in Ω and vanishes everywhere on M , then there exist a neighborhood U of a and holomorphic functions g1 , · · · , gm in U such that f (z) =
m j=1
fj (z)gj (z)
(z ∈ U ).
Several Complex Variables and Integral Formulas
306
Proof. By Theorem 5.10 using a complex linear change of variables there exist a neighborhood U of a = (a′ , an−m+1 , · · · , an ) and holomorphic functions ϕj (w1 , · · · , wn−m ), n − m + 1 ≤ j ≤ n, in a neighborhood U ′ of a′ such that M ∩ U = {w ∈ U | wj = ϕj (w1 , · · · , wn−m ), n − m + 1 ≤ j ≤ n}. If we set ζ1 = wn−m+1 − ϕn−m+1 (w1 , · · · , wn−m ) ···
ζm = wn − ϕn (w1 , · · · , wn−m )
ζm+1 = w1
···
ζn = wn−m ,
then U ∩ M is expressed by U ∩ M = {ζ ∈ U | ζ1 = · · · = ζm = 0}. In case m = 1, we set f (ζ) = ψ1 (ζ). ζ1 Then ψ1 is holomorphic in a neighborhood of 0 and satisfies f (ζ) = ζ1 ψ1 (ζ). This proves Theorem 5.12 in case m = 1. Assume that we have already proved Theorem 5.12 for m−1. Since f is holomorphic in Ω1 = {ζ1 = 0}∩Ω and vanishes in Ω1 ∩ {ζ2 = · · · ζm = 0}, by the inductive hypothesis, there exist holomorphic functions g˜j (ζ2 , · · · , ζn ) (2 ≤ j ≤ m) in a neighborhood of 0 ∈ Cn−1 such that f (ζ) =
m j=2
ζj g˜j (ζ).
The Classical Theory in Several Complex Variables
307
By Theorem 2.14 there exist holomorphic functions gj in a neighborhood W of 0 such that gj = g˜j in {ζ1 = 0} ∩ W . We set m X 1 g1 (ζ) = ζj gj (ζ) . f (ζ) − ζ1 j=2
Then we have the desired equality.
5.3
Oka’s Fundamental Theorem
We give a proof of Oka’s fundamental theorem [OkA2] which is the prototype of the sheaf theory. Moreover, we state the Cartan theorems A and B without giving proofs. Definition 5.9 Let a ∈ Cn . For a neighborhood U of a and f : U → C, we say that (f, U ) is a function element at a. We say that two function elements (f, U ) and (g, V ) at a are equivalent if there exists a neighborhood W ⊂ U ∩ V of a such that f |W = g|W . The equivalence class of a function element (f, U ) at a is called a germ of functions at a and is denoted by fa or γa (f ). Further, we denote by Fa the set of all germs at a. The set of all germs fa such that fa has a representative (f, U ) with f ∈ C(U ) (f ∈ C k (U ), f ∈ O(U )) is denoted by Ca ( Cak , Oa ). By definition we have Oa ⊂ Ca∞ ⊂ Cak ⊂ Ca ⊂ Fa , where k is an integer with 1 ≤ k < ∞. Let (f, U ) and (g, V ) be representatives of fa and ga , respectively. We define fa + ga and fa ga by an equivalent class of (f + g, U ∩ V ) and (f g, U ∩ V ), respectively. Then Fa , Cak and Oa become commutative rings. Definition 5.10 a ∈ Cn , define
For the set Oa of all germs of holomorphic functions at O = OCn =
∪ Oa .
a∈Cn
We introduce the basis of all open sets in O as follows: For an open set U in Cn and a holomorphic function f in U , define Uf = {fz | z ∈ U }.
Several Complex Variables and Integral Formulas
308
We define the basis in O to be the set of all Uf . Define π : O → Cn by π(fa ) = a. Let fa ∈ Oa and let (f, U ) be a representative of fa . Then Uf is a neighborhood of fa and π : Uf → U is bijective, continuous and open mapping, which means that π is a local homeomorphism. Definition 5.11 Let X be a topological space. We say that a topological space S is a sheaf over X if there is a surjective local homeomorphism π : S → X. Hence π is an open mapping. For x ∈ X, Sx = π −1 (x) is called a stalk. Definition 5.12 Let S be a sheaf over X and Y ⊂ X. We say that a continuous mapping s : Y → S is a section of S over Y if π ◦ s(x) = x for all x ∈ Y . We denote by Γ(Y, S) the collection of all sections of S over Y . Definition 5.13 We say that a sheaf S over X is a sheaf of Abelian groups over X if each Sx (x ∈ X) carries the structure of an Abelian group, so that if Y ⊂ X, s1 , s2 ∈ Γ(Y, S), then s1 − s2 : Y → S, being defined by (s1 − s2 )(x) = s1 (x) − s2 (x)
(x ∈ Y ),
is continuous. The sheaf of rings over X is defined similarly. Definition 5.14 Let R be a sheaf of rings over X and let S be a sheaf of Abelian groups over X. We say that S is a sheaf of modules over R (or a sheaf of R-modules), if Sx is a Rx -module, and the product of a section of R and a section of S is a section of S. We say that S is an analytic sheaf if X is a complex manifold and R is a sheaf O of germs of holomorphic functions. Definition 5.15 We say that S ′ ⊂ S is a subsheaf of S if π|S ′ : S ′ → X is a sheaf. Hence S ′ is a subsheaf of S if and only if S ′ is an open subset of S and π(S ′ ) = X. If S is a sheaf of Abelian groups, we assume that Sx′ is a subgroup of Sx . Suppose S ′ and S are sheaves of Aberian groups over X. We say that a continuous mapping ϕ : S ′ → S is a sheaf homomorphism if ϕ(Sx′ ) ⊂ Sx and ϕx = ϕ|Sx′ : Sx′ → Sx is a group homomorphism for each x ∈ X. Definition 5.16
Let ϕ : S ′ → S be a sheaf homomorphism. Define
Ker ϕ = ∪ Ker ϕx , x∈X
Im ϕ = ∪ Im ϕx . x∈X
The Classical Theory in Several Complex Variables
309
Then Ker ϕ is a subsheaf of S ′ and Im ϕ is a subsheaf of S. We have the exact sequence ϕ
ι
0 → Ker ϕ → S ′ → Im ϕ → 0. Definition 5.17 Let S be a sheaf of Abelian groups and let S ′ be a subsheaf of S. Define S/S ′ = ∪ Sx /Sx′ . x∈X
′
We define the quotient sheaf S/S as the union of all the quotient groups Sx /Sx′ for x ∈ X, equipped with the quotient topology, that is, the finest topology which makes the stalkwise defined quotient mapping q : S → S/S ′ continuous. Then q is a sheaf homomorphism and we have the exact sequence ι
q
0 → S ′ → S → S/S ′ → 0. Lemma 5.5 If s1 , s2 ∈ Γ(Y, S) satisfy s1 (x0 ) = s2 (x0 ), then there exists a neighborhood W of x0 such that s1 (x) = s2 (x) for all x ∈ W . Proof. We set s1 (x0 ) = s2 (x0 ) = z0 . Then there exists a neighborhood U of z0 such that π : U → π(U ) is a homeomorphism. We set W = s−1 1 (U ) ∩ (U ). Then W is a neighborhood of x and π ◦ s | = π ◦ s | s−1 0 1 W 2 W = IW , 2 −1 which implies that s1 = s2 = π |W .
Lemma 5.6 Let Ω ⊂ Cn be an open set. For a holomorphic function f in Ω, define sf : Ω → O by sf (a) = fa for a ∈ Ω. Then sf is continuous. Moreover, f ∈ O(Ω) → sf ∈ Γ(Ω, O) is bijective.
Proof. Let a ∈ Ω. For a neighborhood of a, we set Uf = {fz | z ∈ U }. Then we have s−1 f (Uf ) = U , and hence sf is continuous. Since π ◦ s|Ω = IΩ , we have sf ∈ Γ(Ω, O). If f1 = f2 , then there exist z ∈ Ω such that sf1 (z) = sf2 (z), which means that sf1 = sf2 . Next, assume that s ∈ Γ(Ω, O). For a ∈ Ω, we have s(a) = fa . Let (f, U ) be a representative of fa . Since s is continuous, there exists a neighborhood W of a such that s(z) = fz for z ∈ W . Hence there exists f ∈ O(Ω) such that s = sf . Definition 5.18 We say that a commutative ring A with unit is Noetherian if every ideal I ⊂ A is finitely generated, that is, if there exist elements f1 , · · · , fj ∈ I so that every f ∈ I can be written f=
j i=1
ai f i
310
Several Complex Variables and Integral Formulas
for some a1 , · · · , aj ∈ A. Theorem 5.13
O0 is a Noetherian ring.
Proof. In case n = 1, Theorem 5.13 is trivial since every ideal in O0 is generated by a power of z using the Taylor expansion. Assume that Theorem 5.13 has already been proved for the ring O0 (Cn−1 ). Suppose I is an ideal in O0 which contains some non-zero element. Let f ∈ I be a non-zero element. Then by a change of coordinates we may assume that f is regular of order k in zn . For g ∈ I, by the Weierstrass division theorem we have a representation g = qf + r, where r is a polynomial in zn of degree less than k. Let M be a set g ∈ I such that g is a polynomial in zn of degree less than k. Then M is regarded a submodule in O0 (Cn−1 )p . By the inductive hypothesis, O0 (Cn−1 ) is a Noetherian ring, and hence M is finitely generated. If f1 , · · · , fr is the generators for M , then f1 , · · · , fr , f generate I. Consequently, O0 is a Noetherian ring. Lemma 5.7 Let f , g and ω be holomorphic functions in a neighborhood of 0 ∈ Cn . Suppose ω is a Weierstrass polynomial in zn and f is a polynomial in zn . If f = gω, then g is a polynomial in zn . Proof. Since the coefficient of the term of ω of the maximal degree in zn equals 1, f is expressed by f = qω + r, where q and r are polynomials in zn and the degree of r is less than the degree of ω. By the uniqueness of the Weierstrass division theorem we have r = 0, q = g. In order to prove Oka’s fundamental theorem, we need the following lemma. Lemma 5.8 Let {fλ } be a sequence of at most countable non-zero holomorphic functions in a neighborhood U of 0 ∈ Cn . Then there exists an
The Classical Theory in Several Complex Variables
311
invertible linear change of variables zj =
n
αjk zk∗
(j = 1, · · · , n)
k=1
such that for all λ, fλ∗ (z ∗ ) = fλ (z) satisfy the following properties: fλ∗ (z1∗ , 0, · · · , 0) ≡ 0, Proof.
fλ∗ (0, z2∗ , 0, · · · , 0) ≡ 0,
··· ,
fλ∗ (0, · · · , 0, zn∗ ) ≡ 0. (5.3)
In the power series expansion of fλ fλ (z1 , · · · , zn ) =
∞
(λ)
ν1 ···νn =0
aν1 ···νn z1ν1 · · · znνn ,
we rewrite the right side of the above equality by the series of homogeneous polynomials so that sλ is the least homogeneous degree. Since fλ∗ (0, · · · , 0, zk∗ , 0, · · · , 0) =
∞
ν1 ···νn =0
(λ)
aν1 ,··· ,νn (α1k zk∗ )ν1 · · · (αnk zk∗ )νn ,
it is sufficient to choose αjk (j, k = 1, · · · , n) with the following properties: (λ) (k = 1, · · · , n), aν1 ,··· ,νn (α1k )ν1 · · · (αnk )νn = 0 ν1 +···+νn =sλ
det(αjk ) = 0. By the Baire theorem (Lemma 1.4), there exist αjk (j, k = 1, · · · , n) which satisfy the above properties. Theorem 5.14 (Oka’s fundamental theorem) Suppose Ω is an open set in Cn . For z ∈ Ω and F1 , · · · , Fq ∈ O(Ω)p , define a submodule Rz (F1 , · · · , Fq ) of Ozq as follows: Rz (F1 , · · · , Fq ) = {G = (g 1 , · · · , g q ) ∈ Ozq |
q
g j γz (Fj ) = 0}.
j=1
Given z0 ∈ Ω, one can then find a neighborhood ω ⊂ Ω of z0 and finitely many elements G1 , · · · , Gr ∈ O(ω)q such that for any z ∈ ω, Rz (F1 , · · · , Fq ) is generated by γz (G1 ), · · · , γz (Gr ).
Several Complex Variables and Integral Formulas
312
Proof. We assume z0 = 0. (a) Suppose that p > 1. Assume that the theorem has already been proved for p − 1. Let Fj = (Fj1 , · · · , Fjp ). Evidently we have Rz (F1 , · · · , Fq ) ⊂ Rz (F11 , · · · , Fq1 ). By the inductive hypothesis, there exist a neighborhood ω ′ of 0 and H1 , · · · , Hr ∈ O(ω ′ )q such that for any z ∈ ω ′ , Rz (F11 , · · · , Fq1 ) is generated by γz (H1 ), · · · , γz (Hr ). Consequently, we have r (z ∈ ω ′ ). Rz (F1 , · · · , Fq ) ⊂ cj γz (Hj ) | cj ∈ Oz j=1
Let Hj = (Hj1 , · · · , Hjq ). Then if
q r
r
j=1
cj γz (Hj ) ∈ Rz (F1 , · · · , Fq ) if and only
cj γz (Hjk )γz (Fk ) = 0.
(5.4)
k=1 j=1
(5.4) is equivalent to the following equations. q r
cj γz (Hjk Fki ) = 0
k=1 j=1
(i = 1, · · · , p).
(5.5)
Since γz (H1 ), · · · , γz (Hr ) ∈ Rz (F11 , · · · , Fq1 ), we obtain q
γz (Hjk )γz (Fk1 ) =
q
γz (Hjk Fk1 ) = 0
k=1
k=1
(j = 1, · · · , r).
Hence (5.5) holds when i = 1. We set q q k 2 k p Hj Fk , · · · , Hj Fk . Lj = k=1
k=1
r
It follows from (5.5) that j=1 cj γz (Lj ) = 0. By the inductive hypothesis, there exist a neighborhood ω of 0 and K1 , · · · , Ks ∈ O(ω)r such that for z ∈ ω, (c1 , · · · , cr ) is generated by γz (K1 ), · · · , γz (Ks ). Hence there exist αk ∈ Oz for k = 1, · · · , s such that cj =
s
k=1
αs γz (Kkj ).
The Classical Theory in Several Complex Variables
313
Therefore every element of Rz (F1 , · · · , Fq ) has a representation r
j
c γz (Hj ) =
r s
αk γz (Kkj )γz (Hj ),
j=1 k=1
j=1
which implies that if we set Gk =
r j=1
Kkj Hj
(k = 1, · · · , s),
then for z ∈ ω, Rz (F1 , · · · , Fq ) is generated by γz (G1 ), · · · , γz (Gs ). (b) When n = 0, the theorem holds for every p. Assume that the theorem has already been proved for n − 1 dimension and for all p. We will prove the theorem for n dimension and for all p. In (a) we have proved that if the theorem is true for p − 1, then the theorem is true for p when p > 1. Hence it is sufficient to prove the theorem when p = 1. By Lemma 5.8, without loss of generality we may assume that F1 , · · · , Fq are Weierstrass polynomials in zn at 0. We denote by µ the maximum of degrees in zn of F1 , · · · , Fq . We assume that µ is the degree in zn of Fq . We prove the following lemma. Lemma 5.9 Let ζ = (ζ ′ , ζn ) ∈ Cn . Then Rζ (F1 , · · · , Fq ) is generated by finitely many elements whose components are gems of functions in n−1 Oζ ′ [zn ] with a degree in zn which does not exceed µ. Proof of Lemma 5.9 By the Weierstrass preparation theorem we have γζ (Fq ) = F ′ F ′′ , where F ′ is a germ of a Weierstrass polynomial in z − ζ and F ′′ is a germ of holomorphic functions which do not vanish at ζ. Since Fq = znµ + aµ−1 (z ′ )znµ−1 + · · · + a0 (z ′ ), Fq is a polynomial in zn − ζn . By Lemma 5.7, F ′′ is a polynomial in zn − ζn , and hence a polynomial in zn with leading coefficient equal to 1. We denote degrees of F ′ and F ′′ in zn by µ′ and µ′′ , respectively. Let (c1 , · · · , cq ) ∈ Rζ (F1 , · · · , Fq ). Then by the Weierstrass division theorem we have i
ci = t′ F ′ + ri = γζ (Fq )ti + ri
(i = 1, · · · , q − 1),
Several Complex Variables and Integral Formulas
314
i
where t′ , ti , ri ∈ Oζ and each ri is a polynomial in zn with the degree less than µ′ . We set r q = cq −
q−1
γζ (Fi )ti .
i=1
Then we obtain (c1 , · · · , cq ) = γζ (Fq , 0, · · · , 0, −F1 )t1 + γζ (0, Fq , 0, · · · , 0, −F2 )t2
(5.6)
+ · · · + γζ (0, · · · , 0, Fq , −Fq−1 )tq−1 + (r1 , · · · , rq ). Consequently we have (r1 , · · · , rq ) ∈ Rζ (F1 , · · · , Fq ). Hence we have q
q−1
ri γζ (Fi ) =
ri γζ (Fi ) + (rq F ′′ )F ′ = 0.
i=1
i=1
q−1
Since the degree of i=1 ri γζ (Fi ) in zn is less than µ + µ′ , by Lemma 5.7 the degree of rq F ′′ in zn is less than µ. In the equality (r1 , · · · , rq ) =
1 (F ′′ r1 , · · · , F ′′ rq ), F ′′
the degrees of F ′′ rj (j = 1, · · · , q) in zn are less than µ. Hence Lemma 5.9 follows from (5.6), which completes the proof of Lemma 5.9. End of the proof of (b) Let (c1 , · · · , cq ) be one of the elements in Rζ (F1 , · · · , Fq ) described in Lemma 5.9. Then we have j
c =
µ
cjk γζ (znk )
k=0 1
cjk ∈ Oζ ′ .
q
Since (c , · · · , c ) ∈ Rζ (F1 , · · · , Fq ), we have µ q
cjk γζ (znk )γζ (Fj ) = 0.
j=1 k=0
Consequently, we have µ
k=0
(c1k γζ (F1 ) + · · · + cqk γζ (Fq ))γζ (znk ) = 0.
Let Fj (z) = ajµ (z ′ )z µ + ajµ−1 (z ′ )z µ−1 + · · · + aj0 (z ′ ).
(5.7)
The Classical Theory in Several Complex Variables
315
Since coefficients in znk for k = 0, · · · , 2µ in (5.7) are 0, the coefficient of znµ in (5.7) is equal to 0. Hence we have c10 a1µ + · · · + cq0 aqµ + c11 a1µ−1 + · · · +cq1 aqµ−1 + · · · + c1µ a1,0 + · · · + cqµ aq0 = 0. By the inductive hypothesis, there exist a neighborhood ω ′ of 0 ∈ Cn−1 and C1k , · · · , Crk k ∈ O(ω ′ )q , k = 0, · · · , µ, such that (c1k , · · · , cqk ) is generated by C1k , · · · , Crk k . Since (c1 , · · · , cq ) =
µ
k=0
(c1k , · · · , cqk )γζ (znk ),
Rζ (F1 , · · · , Fq ) is generated by germs of C1k znk , · · · , Crk k znk (k = 0, · · · , µ) for ζ = (ζ ′ , ζn ) with ζ ′ ∈ ω ′ . This proves (b). Definition 5.19 An analytic sheaf S on the complex manifold Ω is said to be locally finitely generated if for every z ∈ Ω there exists a neighborhood ω of z and a finite number of sections f1 , · · · , fq ∈ Γ(ω, S) such that Sζ is generated by γζ (f1 ), · · · , γζ (fq ) as an Oζ module for every ζ ∈ ω. Lemma 5.10 Suppose that an analytic sheaf S is locally finitely generated. Let f1 , · · · , fq be sections of S in a neighborhood of z such that γz (f1 ), · · · , γz (fq ) generate Sz . Then γζ (f1 ), · · · , γζ (fq ) generate Sζ for every ζ in a neighborhood of z. Proof. Since S is locally finitely generated, there exist a neighborhood ω of z and g1 , · · · , gr ∈ Γ(ω, S) such that for any ζ ∈ ω, γζ (g1 ), · · · , γζ (gr ) generate Sζ . On the other hand, by the hypothesis we have γz (gi ) =
q
γz (cij )γz (fj )
j=1
(i = 1, · · · , r).
By Lemma 5.5 there exists a neighborhood W of z such that for ζ ∈ W , γζ (gi ) =
q j=1
γζ (cij )γζ (fj )
(i = 1, · · · , r).
Definition 5.20 Let S be an analytic sheaf on the complex manifold Ω and let ω be an open subset of Ω. For f1 , · · · , fq ∈ Γ(ω, S), define the sheaf
Several Complex Variables and Integral Formulas
316
homomorphism h : Oq → S by h
Oq ⊃ Ozq ∋ (g 1 , · · · , g q ) −→
q j=1
g j γz (fj ) ∈ Sz ⊂ S.
q
The subsheaf R(f1 , · · · , fq ) of O is defined by R(f1 , · · · , fq ) = ∪ {(g 1 , · · · , g q ) ∈ Oz | h(g 1 , · · · , g q ) = 0}, z∈ω
and is called the sheaf of relations between f1 , · · · , fq . Definition 5.21 Let S be an analytic sheaf on the complex manifold Ω. S is called coherent if (1) S is locally finitely generated. (2) If ω is an open subset of Ω and f1 , · · · , fq ∈ Γ(ω, S), then the sheaf of relations R(f1 , · · · , fq ) is locally finitely generated. Theorem 5.15 ent.
Every locally finitely generated subsheaf of Op is coher-
Proof. We show (2) in the definition of the coherent sheaf. Since f1 , · · · , fq ∈ O(ω)p , by Oka’s fundamental theorem (Theorem 5.14) the sheaf of relations R(f1 , · · · , fq ) is locally finitely generated. Theorem 5.16 Let S be a coherent sheaf on the complex manifold Ω and let ω be an open subset of Ω. If f1 , · · · , fq ∈ Γ(ω, S), then the sheaf of relations R(f1 , · · · , fq ) is also coherent. Proof. Since S is coherent, R(f1 , · · · , fq ) is locally finitely generated. Theorem 5.16 follows from Theorem 5.15 and the fact that R(f1 , · · · , fq ) is a subsheaf of Oq . Example 5.1 Proof. Define
There is a subsheaf of O which is not coherent.
Suppose that ω and Ω are open sets in C with φ = ω ⊂ Ω, ω = Ω. Sz =
Oz (z ∈ ω) . 0 (z ∈ Ω\ω)
Then S is a subsheaf of O. Every section of S over a connected open set which intersects Ω\ω must be 0 (see Exercise 1.5). Hence if S is finitely generated in some connected neighborhood of a boundary point of ω, then we have Sz = 0 in the neighborhood, which is a contradiction.
The Classical Theory in Several Complex Variables
317
Definition 5.22 Let X be a paracompact Hausdorff space and let S be a sheaf of Abelian groups in X. Let U = {Uj | j ∈ J} be an open cover of X and let q be a nonnegative integer. We say that c is a qcochain for U with coefficients in S if c is a mapping which assigns to each (q + 1) tuple (j0 , j1 , · · · , jq ) ∈ J q+1 with Uj0 ∩ · · · ∩ Ujq = φ a section cj0 ,j1 ,··· ,jq ∈ Γ(Uj0 ∩ · · · ∩ Ujq , S). Define cj0 ,··· ,jq = εci0 ,··· ,iq , where ε = ±1 is a sign of the permutation j0 , j1 , · · · , jq . i0 , i1 , · · · , iq We denote by C q (U, S) the set of all q-cochains. A coboundary mapping δq : C q (U, S) → C q+1 (U, S) is defined as follows: (δq c)j0 ···jq+1 =
q+1
k=0
(−1)k cj0 ···ˆjk ···jq+1 ,
where j0 · · · ˆjk · · · jq+1 means that jk is omitted. By definition we have δq+1 ◦ δq = 0
(q ≥ 0).
(5.8)
Define Z q (U, S) = {c ∈ C q (U, S) | δq c = 0}
(q ≥ 0)
B q (U, S) = {δq−1 c | c ∈ C q−1 (U, S)}
(q ≥ 1).
and
An element Z q (U, S) is called a q-cocycle and an element of B q (U, S) is called a q-coboundary. Define B 0 (U, S) = 0. It follows from (5.8) that B q (U, S) ⊂ Z q (U, S). Define H q (U, S) := Z q (U, S)/B q (U, S). ˘ H q (U, S) is called a q-th Cech cohomology group of U with coefficients in S. Definition 5.23 Let V = {Vi | i ∈ I} be a refinement of U, that is, there exists a mapping τ : I → J such that Vi ⊂ Uτ (i) for i ∈ I. Define τq∗ : C q (U, S) → C q (V, S)
(q ≥ 0)
by (τq∗ (c))i0 ···iq = cτ (i0 )···τ (iq ) |Vi0 ∩···∩Viq .
Several Complex Variables and Integral Formulas
318
∗ Since τq+1 ◦ δq = δq ◦ τq∗ , we define V ρU : H q (U, S) → Hq (V, S) q
by V ∗ ρU q ([c]) = [τq c].
We have the following lemma. We omit the proof. V ρU is independent of the choice of τ . q
Lemma 5.11
Definition 5.24 For two open covers U, W of X, we say that [c] ∈ H q (U, S) and [d] ∈ H q (W, S) are equivalent if there exists a refinement V of U and W such that WV V ρU ([d]). q ([c]) = ρq
We denote by H q (X, S) the set of all equivalent classes by this equivalent ˘ relation. H q (X, S) is an Abelian group. H q (X, S) is called the q-th Cech cohomology group of X with coefficients in S. By definition we have the following lemma. Lemma 5.12
For an open cover U of X, we have H 0 (X, S) = H 0 (U, S) = Z 0 (U, S) = Γ(X, S).
Definition 5.25 Let ϕ : S ′ → S be a sheaf homomorphism and U an open cover of X. Define ϕ : C q (U, S ′ ) → C q (U, S ′ ) by ϕ(c) = ϕ ◦ c. Moreover, we define ϕqU and ϕq using ϕ such that ϕqU : H q (U, S ′ ) → H q (U, S) and ϕq : H q (X, S ′ ) → H q (X, S). Then we have the following theorem. Theorem 5.17
Suppose ϕ
ψ
0 → S ′ → S → S ′′ → 0
The Classical Theory in Several Complex Variables
319
is an exact sequence of sheaf homomorphisms over X and H 1 (X, S ′ ) = 0. Then ψ 0 : Γ(X, S) → Γ(X, S ′′ ) is surjective. Proof. Let s′′ ∈ Γ(X, S ′′ ). For x ∈ X, we have s′′ (x) ∈ Sx′′ . Since ψ is surjective, there exists sx ∈ Sx such that ψ(sx ) = s′′ (x). There exist a neighborhood W of x and a section sˆx ∈ Γ(W, S) such that sˆx (x) = sx (see Exercise 5.2). Consequently, we have ψ ◦ sˆx (x) = s′′ (x). By Lemma 5.5 there exists a neighborhood Ux ⊂ W of x such that ψ ◦ sˆx = s′′ in Ux . We set U = {Ux | x ∈ X}. Then we have sˆ = {sx } ∈ C 0 (U, S). Since ψ ◦ (δ0 sˆ) = δ0 (ψ ◦ sˆ) = δ0 s′′ = 0, we have δ0 sˆ ∈ B(U, Kerψ). Since Im ϕ = Ker ψ, there exist a refinement V = {Vx | x ∈ X} (Vx ⊂ Ux ) of U and s′ ∈ C 1 (V, S ′ ) such that ϕ ◦ s′ = δ0 sˆ. Consequently, ϕ ◦ (δ1 s′ ) = δ1 (ϕ ◦ s′ ) = δ1 (δ0 sˆ) = 0. Since ϕ is injective, we have δ1 s′ = 0, which implies that s′ ∈ Z 1 (V, S ′ ). Since H 1 (X, S ′ ) = 0, taking a refinement of V if necessary, we may assume that s′ ∈ B 1 (V, S ′ ). Hence there exists g ∈ C 0 (V, S ′ ) such that s′ = δ0 g. Thus we have δ0 sˆ = ϕ ◦ s′ = ϕ ◦ (δ0 g) = δ0 (ϕ ◦ g). If we set s = sˆ − ϕ ◦ g, then we have δ0 s = 0, and hence s ∈ Γ(X, S). Moreover, we have ψ ◦ s = ψ ◦ sˆ − ψ ◦ ϕ ◦ g = s′′ , which means that ψ 0 (s) = s′′ . Hence ψ 0 is surjective. Definition 5.26 manifold if
A σ compact complex manifold Ω is said to be a Stein
(a) Ω is holomorphically convex, that is, for any compact subset K of Ω, ˆ O = {z ∈ Ω | |f (z)| ≤ sup |f | for all f ∈ O(Ω)} K Ω K
is compact.
Several Complex Variables and Integral Formulas
320
(b) For z1 , z2 ∈ Ω z1 = z2 , there exists f ∈ O(Ω) such that f (z1 ) = f (z2 ). (c) For every z ∈ Ω, one can find n functions f1 , · · · , fn ∈ O(Ω) which form a coordinate system at z. Remark 5.1 Every pseudoconvex domain in Cn satisfies (a), (b) and (c), and hence it is a Stein manifold. Theorem 5.18 fold.
Every submanifold of a Stein manifold is a Stein mani-
Proof. Let V be a submanifold of a Stein manifold Ω. Since O(Ω) ⊂ ˆ O . Hence V is holomorphically convex. This ˆO ⊂ K O(V ), we have K V Ω proves (a). (b) is trivial. Let v ∈ V . Then there exist a neighborhood ω of v and a local coordinate system z1 , · · · , zn in ω such that ω ∩ V = {w ∈ ω | zm+1 (w) = · · · = zn (w) = 0}. Let f1 , · · · , fn ∈ O(Ω) be a coordinate system at v. Then at z(v) we have ∂fi det = 0 (i, j = 1, · · · , n). ∂zj Hence we can choose i1 , · · · , im such that ∂fiµ det = 0 (µ, j = 1, · · · , m). ∂zj Thus the restrictions of fi1 , · · · , fim to V form a local coordinate system at v. Definition 5.27 We say that a subset A of a Stein manifold Ω is an analytic subset if A is a closed subset of Ω and for any p ∈ A there exist a neighborhood Up of p and holomorphic functions h1 , · · · , hkp in Up such that Up ∩ A = {z ∈ Up | h1 (z) = · · · = hkp (z) = 0}. Definition 5.28 Let A be an analytic subset of a Stein manifold Ω. A continuous function f : A → C is said to be holomorphic in A if for any a ∈ A there exist a neighborhood Ua of a in Ω and a holomorphic function ha in Ua such that f (z) = ha (z) for all z ∈ A ∩ Ua . Definition 5.29 Let A be an analytic subset of a Stein manifold Ω. We define a subsheaf FA of O in such a way that (FA )z = Oz for z ∈ A , and (FA )z = {fz ∈ Oz | f |A = 0} for z ∈ A. FA is called the sheaf of ideals of the analytic subset A.
The Classical Theory in Several Complex Variables
321
We have the following theorem. The proof is omitted (see GunningRossi [GUR]). Theorem 5.19 Every sheaf of ideals I of an analytic subset of a Stein manifold is coherent. We omit the proof of the following lemma (see Gunning-Rossi [GUR]). Lemma 5.13
If in an exact sequence of sheaves 0 → S ′ → S → S ′′ → 0
any two of the sheaves S ′ , S, S ′′ are coherent sheaves, then the third is also coherent. The following theorem is known as Theorem A and Theorem B of Cartan. We omit the proof (see H¨ ormander [HR2], Gunning-Rossi [GUR]). Theorem 5.20 Ω. Then
Let Ω be a Stein manifold and A a coherent sheaf over
(a) (Cartan theorem A) Let z ∈ Ω. For any s ∈ Az there exist f1 , · · · , fk ∈ Γ(Ω, A) and s1 , · · · , sk ∈ Oz such that s=
k
sj (fj )z .
j=1
(b) (Cartan theorem B) H q (Ω, A) = 0
(q ≥ 1).
Corollary 5.5 Let Ω be a Stein manifold and let A be an analytic subset of Ω. Then for any z ∈ A there exists f ∈ O(Ω) such that f (z) = 0, f |A = 0. Proof. Let z ∈ A. Then there exist a neighborhood U of z and a holomorphic function s in U such that s(z) = 0, s|A = 0. Let FA be the sheaf of ideals of A. By the Cartan theorem A, there exist holomorphic functions s1 , · · · , sk in a neighborhood of z and f1 , · · · , fk ∈ Γ(Ω, FA ) such that s=
k
sj (fj )z .
j=1
Hence there exists j0 with 1 ≤ j0 ≤ k such that fj0 (z) = 0.
Corollary 5.6 Let Ω be a Stein manifold and let A be an analytic subset of Ω. Then every holomorphic function in A can be extended to a holomorphic function in Ω.
322
Several Complex Variables and Integral Formulas
Proof. We denote by FA the sheaf of ideals of A. Γ(Ω, O(Ω)/FA ) can be regarded as the set of all holomorphic functions in A. Since FA is coherent, we have H 1 (Ω, FA ) = 0. By the exact sequence of sheaves 0 → FA → O(Ω) → O(Ω)/FA → 0, Γ(Ω, O(Ω)) → Γ(Ω, O(Ω)/FA ) is surjective, which means that for a holo morphic function f in A there exists F ∈ Γ(Ω, O(Ω)) with F |A = f . Definition 5.30 Let X be a complex manifold. An open cover U = {Ui }i∈I of X is said to be a Stein cover if U is a locally finite cover and each Ui is a Stein open set. The following theorem holds. We omit the proof (see Grauert and Remmert [GRR]). Theorem 5.21 Let X be a complex manifold and let U be a Stein cover of X, S a coherent sheaf over X. Then H q (U, S) = H q (X, S)
(q ≥ 0).
The following theorem follows from Theorem 5.19, Theorem 5.20 and Theorem 5.21. Theorem 5.22 Let Ω ⊂ Cn be a pseudoconvex domain. Let A be an analytic subset of Ω and let FA be the sheaf of ideals of A, {Uj }j∈I a Stein cover of Ω. Suppose fij ∈ Γ(Ui ∩ Uj , FA ) satisfy the equalities fij (z) + fjk (z) + fki (z) = 0
(z ∈ Ui ∩ Uj ∩ Uk , i, j, k ∈ I).
Then there exist fj ∈ Γ(Uj , FA ) such that fij (z) = fi (z) − fj (z)
(z ∈ Ui ∩ Uj , i, j ∈ I).
Theorem 5.23 Let A be a coherent analytic sheaf over a Stein manifold Ω and let S be a subsheaf of A. Let s1 , · · · , sk ∈ Γ(Ω, A). Suppose that s1 , · · · , sk generate Sz over Oz for each z ∈ Ω. Then for s ∈ Γ(Ω, S), there exist f1 , · · · , fk ∈ Γ(Ω, O) such that s=
k j=1
f j sj .
The Classical Theory in Several Complex Variables
Proof.
323
Define ϕ : O(Ω)k → A by ϕ(b1 , · · · , bk ) =
k
b j sj
j=1
((b1 , · · · , bk ) ∈ Ozk , z ∈ Ω).
By Theorem 5.16 Ker ϕ is coherent. By the Cartan theorem B, we have H 1 (Ω, Ker ϕ) = 0. By applying Theorem 5.18 to the exact sequence of sheaves ϕ
0 → Ker ϕ → O(Ω)k → S → 0 we have that ϕ0 : Γ(Ω, O(Ω)k ) → Γ(Ω, S) is surjective. Consequently, for s ∈ Γ(Ω, S), there exist (f1 , · · · , fk ) ∈ Γ(Ω, O(Ω)k ) such that s=
k
f j sj .
j=1
Corollary 5.7 Let Ω be a pseudoconvex domain in Cn and let A be an analytic subset of Ω. Suppose that there exist holomorphic functions s1 (z), · · · , sk (z), k ≤ n, in Ω such that A = {z ∈ Ω | s1 (z) = · · · = sk (z) = 0} and F = (s1 , · · · , sk ) is nonsingular in Ω. If g is a holomorphic function in Ω with g|A = 0, then there exist holomorphic functions f1 , · · · , fk in Ω such that g(z) =
k j=1
fj (z)sj (z)
(z ∈ Ω).
Proof. Let I be the sheaf of ideals of A. We apply Theorem 5.23 by setting A = O(Ω), S = I. We have g ∈ Γ(Ω, I). By Theorem 5.12 (s1 , · · · , sk ) satisfies the hypothesis of Theorem 5.23. Theorem 5.24 Let Ω be a Stein manifold and let K be a compact subset ˆ Then there exists ϕ ∈ C ∞ (Ω) with the of Ω, ω a neighborhood of K. following properties: (a) ϕ is a strictly plurisubharmonic function in Ω. (b) ϕ < 0 in K and ϕ > 0 in Ω\ω. (c) For every c ∈ R, {z ∈ Ω | ϕ(z) < c} ⊂⊂ Ω.
Several Complex Variables and Integral Formulas
324
ˆ instead K ˆ O . Since Ω is Proof. For simplicity, we adopt the notation K Ω σ compact, there exists a sequence {Kj } of compact sets such that Ω = ∞ ˆ ˆ ˆ ∪∞ j=1 Kj , Kj ⊂ Kj+1 . Hence we have Ω = ∪j=1 Kj , Kj ⊂ Kj+1 . By ˆ j , we can obtain a sequence {Kj } of compact subsets of replacing Kj by K Ω such that ∞
◦ ˆ Kj ⊂ Kj+1 ˆ j = Kj , Ω = ∪ Kj . K1 = K, , K j=1
We choose open sets ωj with the properties that Kj ⊂ ωj ⊂ Kj+1 , ω1 ⊂ ω. Let a ∈ Kj+2 − ωj . Since a 6∈ Kj , there exists fja ∈ O(Ω) such that |fja (a)| > sup |fja |. Kj
We choose αja such that |fja (a)| > αja > sup |fja |. Kj
We set gja = fja /αja . Then we have sup |gja | < 1.
|gja (a)| > 1,
Kj
Since Kj+2 − ωj is a compact set, there exist an open set Wjk and functions gjk ∈ O(Ω), k = 1, · · · , kj , such that kj
Kj+2 − ωj ⊂ ∪ Wjk , sup |gjk | < 1, k=1
Kj
|gjk (z)| > 1 (z ∈ Wjk ).
Consequently, we have sup |gjk | < 1 (k = 1, · · · , kj ), Kj
max |gjk (z)| > 1 (z ∈ Kj+2 − ωj ). k
m (m is sufficiently large), we obtain Replacing gjk by gjk kj X
|gjk (z)|2
j
(z ∈ Kj+2 − ωj ).
(5.10)
The Classical Theory in Several Complex Variables
325
Further we may assume that gjk , k = 1, · · · , kj , contains n functions which form the coordinate system at any point in Kj . Define ϕ(z) =
kj ∞ j=1 k=1
|fjk (z)|2 − 1.
(5.11)
By (5.9) the series in the right side of (5.11) converges. By (5.10) we have ϕ > j − 1 in ωjc . Therefore we have ϕ > 0 in ω c . By (5.9) we have ∞ ϕ < j=1 2−j = 1 in K. On the other hand kj ∞
fjk (z)fjk (ζ)
j=1 k=1
converges uniformly on every compact subset of Ω × Ω and is holomorphic ¯ and hence can be expanded to a power series. Hence we obtain in (z, ζ), 2 kj n ∞ n ∂2ϕ ∂fjk (z)ws w ¯t = (z)ws . ∂zs ∂ z¯t ∂zs s,t=1
j=1 k=1 s=1
Assume that for all j, k
n ∂fjk s=1
∂zs
(z)ws = 0.
Since fjk (k = 1, · · · , kj ) contain n functions which form a coordinate system at z, we have w = 0. Hence ϕ is strictly plurisubharmonic. (c) is trivial, which completes the proof of Theorem 5.24. Lemma 5.14 Let Ω be a Stein manifold and let K be a compact subset ˆ O . Let ω be a neighborhood of K. Then there exists an of Ω with K = K Ω analytic polyhedron P such that K ⊂ P ⊂⊂ ω. ˆ O , there Proof. We may assume that ω ⊂⊂ Ω. Let z ∈ ∂ω. Since z ∈ K Ω exists f ∈ O(Ω) such that |f (z)| > sup |f |. K
We choose α such that |f (z)| > α > supK |f |. We set g = f /α. Then |g| < 1 in K, |g(z)| > 1. By the Heine-Borel theorem, there exist f1 , · · · , fN ∈ O(Ω) such that if we set P ′ = {z ∈ Ω | |gj (z)| < 1 (j = 1, · · · , N )},
Several Complex Variables and Integral Formulas
326
then K ⊂ P ′ , ∂ω ∩ P ′ = φ. Let P = ω ∩ P ′ . Then P is an analytic polyhedron we seek. 5.4
The Cousin Problem
We study the Cousin problem using the L2 estimate for solutions of the ∂¯ problem on Stein manifolds due to H¨ormander [HR2]. H¨ormander [HR2] proved the following theorem. We omit the proof. ∞ (Ω) Theorem 5.25 Let Ω be a Stein manifold. Then for f ∈ C(p,q+1) ∞ ¯ ¯ with ∂f = 0, there exists u ∈ C(p,q) (Ω) such that ∂u = f .
Theorem 5.26 (First Cousin problem) Let Ω be a Stein manifold and let {ωj } be a sequence of open subsets of Ω with Ω = ∪∞ j=1 ωj . Suppose that gjk ∈ O(ωj ∩ ωk ), j, k = 1, 2, · · · , satisfy the following conditions: (a) gjk = −gkj . (b) gij + gjk + gki = 0 in ωi ∩ ωj ∩ ωk . Then there exist gj ∈ O(ωj ) such that gjk = gk − gj in ωj ∩ ωk Proof. Let {ϕν } be a partition of unity subordinate to {ωj }. Then we have ϕν ∈ Cc∞ (ωiν ). Further, for any compact subset K of Ω, ϕν equals identically zero on K except for a finite number of ν and ∞
ϕν = 1.
ν=1
Suppose that gjk is expressed by gjk = gk − gj in ωj ∩ ωk . Let j = iν . Then we have giν k = gk − giν . If we multiply by ϕν and add with respect to ν, then we obtain ∞
ν=1
ϕν giν k = gk −
∞
ν=1
ϕν giν .
The Classical Theory in Several Complex Variables
327
Define hk =
∞
ϕν giν k .
ν=1
Then we have hk ∈ C ∞ (ωk ). Moreover we have hk − hj =
∞
ν=1
ϕν (giν k − giν j ) =
∞
ϕν gjk = gjk
ν=1
¯ k in ωk , then ψ ∈ C ∞ (Ω) and ∂ψ ¯ = 0. in ωj ∩ ωk . If we set ψ = ∂h (0,1) ¯ = −ψ. We set By Theorem 5.25, there exist u ∈ C ∞ (Ω) such that ∂u gk = hk + u. Then gk are solutions we seek. Lemma 5.15 Let Ω ⊂ RN be a simply connected domain. Suppose f : Ω → C is continuous and nowhere vanishing. Then there exists a continuous function g in Ω such that f = eg . Proof. Let P ∈ Ω. Without loss of generality we may assume that Re f (P ) > 0. Then there exists a neighborhood UP of P such that f (UP ) ⊂ {z | Rez > 0}. Hence we can define a continuous function log f in UP . Fix P0 ∈ Ω. Let γ : [0, 1] → Ω be a smooth Jordan closed curve such that γ(0) = γ(1) = P0 . For each P on γ we can choose a neighborhood Up of P having the property mentioned above. Then we can define a function log f (γ(t)) for 0 ≤ t < 1. Assume that log f ◦ γ(0) = limt→1− log f◦ γ(t). Since Ω is simply connected, there exists a continuous function u(s, t) on [0, 1] × [0, 1] such that u(0, t) = γ(t) u(s, 0) = u(s, 1) = P0 u(1, t) = P0
(0 ≤ t ≤ 1), (0 ≤ s ≤ 1), (0 ≤ t ≤ 1).
If we set ρ(s) =
1 { lim log f (u(s, t)) − log f (u(s, 0))}, 2πi t→1−
then ρ(s) is an integer valued continuous function and equals 0 when s is close to 1. Then ρ(0) = 0, which is a contradiction. Hence we can define log f (γ(t)) for 0 ≤ t ≤ 1. Since γ is an arbitrary closed Jordan curve, we can define log f in Ω. We set g = log f . Then f = eg .
Several Complex Variables and Integral Formulas
328
Lemma 5.16 Let Ω ⊂ Cn be a simply connected domain and let f : Ω → C be holomorphic and nowhere vanishing. Then there exists a holomorphic function g in Ω such that f = eg . Proof. By Lemma 5.15 there exists a continuous function g in Ω such that f = eg . Then we have in the sense of distributions 0=
∂g ∂f = eg , ∂ z¯j ∂ z¯j
and hence ∂g = 0. ∂ z¯j Hence g is holomorphic.
Definition 5.31 We denote by O∗ (Ω) the set of all nowhere vanishing holomorphic functions in a complex manifold Ω. We also denote by C ∗ (Ω) the set of all nowhere vanishing continuous functions on a complex manifold Ω. Theorem 5.27 (Second Cousin problem) Let Ω be a Stein manifold and let {ωj } be a sequence of open subsets of Ω with Ω = ∪∞ j=1 ωj . Suppose ∗ that gjk ∈ O (ω ∩ ωk ), j, k = 1, 2, · · · , satisfy the following properties: (a) gjk gkj = 1. (b) gij gjk gki = 1 in ωi ∩ ωj ∩ ωk . Moreover, suppose there exist cj ∈ C ∗ (ωj ) with the properties gjk = ck c−1 j in ωj ∩ ωk . Then there exist gj ∈ O∗ (ωj ) such that gjk = gk gj−1 in ωj ∩ ωk . Proof.
By the assumption there exist cj ∈ C ∗ (ωj ) such that gjk = ck c−1 j
in ωj ∩ ωk . First we assume that ωj is simply connected. By Lemma 5.16 there exist bj ∈ C(Ω) such that cj = ebj . We set hjk = bk − bj . Then we have hjk . gjk = ck c−1 j =e
The Classical Theory in Several Complex Variables
329
Using the same method as in the proof of Lemma 5.16, hjk is holomorphic in ωj ∩ ωk . Evidently we have hij = −hji ,
hij + hjk + hki = 0.
By Theorem 5.26 there exist hk ∈ O(ωk ) such that hjk = hk − hj in ωj ∩ ωk . We set gk = ehk . Then gk gj−1 = gjk . Next we prove the general case. Let {ων′ } be a refinement of {ωj } whose elements are simply connected open subsets of Ω. Then for any ν, there exists iν such that ων′ ⊂ ωiν . Define ′ gνµ = giν iµ
in ων′ ∩ ωµ′ . Then from the proof of the first half, there exist gµ′ ∈ O∗ (ωµ′ ) such that ′ = gµ′ gν′ gνµ
−1
.
in ων′ ∩ ωµ′ . Consequently, we obtain gµ′ gν′
−1
giµ i giiν = 1
in ωi ∩ων′ ∩ωµ′ ⊂ ωi ∩ωiν ∩ωiµ . Hence we have gµ′ giµ i = gν′ giν i in ωi ∩ων′ ∩ωµ′ . If we define gi = gν′ giν i in ωi ∩ ων′ , then gi ∈ O∗ (ωi ). Therefore we obtain gk gj−1 = gν′ giν k (gν′ giν j )−1 = giν k gjiν = gjk in ων′ ∩ ωj ∩ ωk .
Exercises 5.1 (Poincar´ e theorem) Define ∆ = {z ∈ C | |z| < 1},
B = {w ∈ C2 | |w| < 1}.
Several Complex Variables and Integral Formulas
330
Show that there is no biholomorphic mapping F = (f1 , f2 ) : ∆×∆ → B by proving the following: (a) For w ∈ ∆, define a holomorphic mapping Fw : ∆ → B by ∂f2 ∂f1 (z, w), (z, w) . Fw (z) = ∂w ∂w Then for any z0 ∈ ∂∆ we have lim Fw (z) = 0.
z→z0
(b) F (z, w) is constant with respect to w. 5.2 5.3
Let (S, π, X) be a sheaf over X. Show that if sx ∈ Sx , then there exists a neighborhood U of x and s ∈ Γ(U, S) such that s(x) = sx .
Suppose C 1 curve ϕ : [0, 2π] → C\{0} satisfies ϕ(0) = ϕ(2π). Show that 2π ′ ϕ (θ) 1 dθ N (ϕ) = 2πi 0 ϕ(θ)
is an integer. 5.4
Suppose g is a C 1 function in {z ∈ C | |z| ≤ 1} and nowhere vanishing. Prove that if we set ϕ(θ) = g(eiθ ), then N (ϕ) = 0.
5.5 (Oka’s counterexample)
Define 3 4
Ω = {(z1 , z2 ) ∈ C2 |
< |zj |
0}
and U1 = Ω − ω1 ,
U2 = Ω − ω 2 .
Show that (a) A ∩ {z ∈ Ω | Im z1 = 0} = φ,
A = ω1 ∪ ω2 ,
Ω = U1 ∪ U2 .
(b) Define f1 = z2 − z1 + 1 in U1 , f2 = 1 in U2 . Then f2 f1−1 ∈ O∗ (U1 ∩ U2 ). (c) There is no f ∈ O(Ω) which satisfies f /f2 ∈ O∗ (U2 ), f /f1 ∈ O∗ (U1 ).
Appendix A
Compact Operators
In Appendix A we prove Proposition A.10 and Proposition A.13 concerning compact operators which are needed to prove Theorem 3.30 and Theorem 3.29, respectively. For the proofs we refer to Berezansky-Sheftel-Us [BES]. Let E1 and E2 be normed spaces and let A : E1 → E2 be a bounded operator. Define A∗ : E2′ → E1′ by (A∗ f )(x) = f (Ax)
(f ∈ E2′ , x ∈ E1 ).
(A.1)
A∗ is called a conjugate operator of A. For a bounded linear operator A : E1 → E2 , A∗ : E2′ → E1′ is a bounded linear operator. Moreover, we have A∗ = A. Let X and Y be normed spaces. (1) We denote by B(X, Y ) the set of all bounded linear operators T : X → Y . Moreover, we denote B(X, X) by B(X). (2) A linear operator T : X → Y is called a compact operator if for any bounded sequence {xn } of X, there exists a subsequence{xni } of {xn } such that {T (xni )} converges to a point in Y . (3) We say that T ∈ B(X, Y ) is invertible if there exists S ∈ B(Y, X) such that ST = IX ,
T S = IY ,
where IX is the identity mapping from X onto X and IY is the identity mapping from Y onto Y . In this case we write S = T −1 . Proposition A.1 (Ascoli-Arzela theorem) Let X be a compact topological space and let C(X) be a Banach space consisting of all continuous functions on X. That is, if we define the metric for f, g ∈ C(X) by d(f, g) = f − g = sup{|f (x) − g(x)| | x ∈ X}, 331
Several Complex Variables and Integral Formulas
332
then C(X) is a complete metric space). Suppose Φ ⊂ C(X) satisfies the following properties: (a) sup{|f (x)| | x ∈ X, f ∈ Φ} = M < ∞. (b) For any ε > 0 and any x ∈ X there exist a neighborhood V such that |f (y) − f (x)| < ε
(y ∈ V, f ∈ Φ).
Then every sequence {fn } in Φ contains a convergent subsequence. Proof. Since X is compact, for any positive integer k, it follows from (b) that there exist a finite subset Fk of X and a neighborhood Vyk of y ∈ Fk such that X = ∪ Vyk y∈Fk
and |f (x) − f (y)|
0 such that there are infinitely many n with |λn | ≤ N . Hence we can choose a convergent subsequence {λkn } of {λn }. We set lim λkn = λ. Taking into n→∞ account that xkn − (z − λy) ≤ xkn + λkn y − z + λkn y − λy → 0, we have lim xkn = z − λy. If we set lim xkn = x, then x ∈ V and n→∞ n→∞ x = z − λy, and hence z ∈ V ∗ . Hence V ∗ is closed. Proposition A.4 Let E be a normed space and let G be a closed subspace of E with E = G. Then for any ε > 0 there exists yε ∈ G such that yε = 1,
yε − x > 1 − ε
(x ∈ G).
Proof. Let z ∈ G. Since G is closed, δ = ρ(z, G) = inf{z − x | x ∈ G} > 0. By the definition of the infimum, for any η > 0 there exists xη ∈ G such that δ ≤ z − xη < δ + η. We choose η such that ε = η(δ + η)−1 and set yε = z − xη −1 (z − xη ). We show that yε satisfies the desired properties. Clearly we have yε ∈ G, yε = 1. Let x ∈ G. Then we have yε − x = z − xη −1 z − (xη + xz − xη ). Taking into account that xη + xz − xη ∈ G, we obtain yε − x ≥ z − xη −1 δ >
δ η =1− = 1 − ε. δ+η δ+η
Proposition A.5 Let E be a normed space. If every bounded sequence in E contains a convergent subsequence, then E is a finite dimensional space.
Compact Operators
335
Proof. Suppose E is an infinite dimensional space. Let x1 ∈ E be such that kx1 k = 1. We set G1 = {λx1 | λ ∈ F }, where F is the set of all scalars. It follows from Theorem A3 that there exists x2 6∈ G1 such that kx1 k = 1,
kx2 − xk >
1 2
(x ∈ G1 ).
In particular, we have kx2 − x1 k > 21 . Let G2 be a vector space generated by x1 , x2 . By Theorem A3 there exists x3 6∈ G2 such that kx3 k = 1
kx3 − xk >
1 2
(x ∈ G2 ).
In particular, we have kx3 − x2 k >
1 , 2
kx3 − x1 k >
1 . 2
Repeating this process, there exists a sequence {xn } such that kxn k = 1, kxn − xm k > 21 for m 6= n. Then {xn } does not contain any convergent subsequence, which contradicts the hypothesis. Hence E is a finite dimensional space. Proposition A.6 Let E be a Banach space and let A : E → E be a compact operator, T = A − I, where I: E → E is the identity mapping. Then Ker T = {x ∈ E | T x = 0} is a finite dimensional space. Proof. Let {xn } be a bounded sequence in Ker T . Since A(xn ) = xn , {xn } contains a convergent subsequence. By Theorem A4, Ker T is a finite dimensional space. Proposition A.7 Let E be a Banach space and let A : E → E be a compact operator, T = A − I. Then T (E) is a closed subspace of E. Proof. T (E) is a vector space. We show that T (E) is a closed subset of E. First we show that there is a constant c > 0 depending only on T such that for y ∈ T (E) there exists x with T x = y,
kxk ≤ ckyk.
(A.2)
Suppose x0 satisfies T x0 = y. Then any solution x of the equation T x = y can be written x = x0 + z, where z is a solution of T z = 0. Hence we have d := inf{kxk | T x = y} = inf{kx0 + zk | z ∈ Ker T }. Then there exists a sequence {zn } ⊂ Ker T such that kx0 + zn k → d. Consequently, {zn } is bounded. Since Ker T is a finite dimensional space in
336
Several Complex Variables and Integral Formulas
view of Proposition A.6, we have a representation zn = an1 x1 + · · · + ank xk , where {x1 , · · · , xk } is a basis of Ker T . Suppose {an1 } is not bounded. Then there exists a subsequence {an1 i } of {an1 } such that lim an1 i = ∞. We set i→∞
anj
αnj = qP k
.
n 2 i=1 (ai )
ni i Then |αnj | ≤ 1. We can choose a convergent subsequence {αm j } of {αj }. Since we have )1/2 ( k X mi mi 2 i (αm (ai ) z mi = 1 x1 + · · · + αk xk ), i=1
which implies that lim kzmi k = ∞. This contradicts the hypothesis. Hence i→∞
{an1 } is bounded, and hence {an1 } contains a convergent subsequence, which means that {zn } contains a convergent subsequence {znk }. We set z0 = lim znk . k→∞
Then we have kx0 + z0 k = lim kx0 + znk k = d. k→∞
We set x ˆ = x0 + z0 . Then T (ˆ x) = y. Now we show that x ˆ satisfies (A.2). Suppose (A.2) does not hold. For any positive integer n there exists yn ∈ T (E) such that kˆ xn k > nkyn k,
T (ˆ xn ) = yn .
(A.3)
We set ξˆn = kˆ xn k−1 x ˆn ,
ηn = kˆ xn k−1 yn .
If T ξ = ηn , then kξk ≥ 1. Hence ξˆn has the smallest norm among solutions of the equation T ξ = ηn . Since {ξˆn } is bounded, {Aξˆn } contains a convergent subsequence {A(ξˆnk )}. We set ξ0 = lim A(ξˆnk ). By (A.3) we have k→∞
lim ηn = 0. Taking into account that A(ξˆnk ) − ξˆnk = ηnk , we have
n→∞
ξ0 = lim ξˆnk . k→∞
Since A is continuous, we have A(ξ0 ) = lim A(ξˆnk ). k→∞
Compact Operators
337
We obtain A(ξ0 ) = ξ0 , and hence ξ0 ∈ Ker T . Thus we have T (ξˆnk − ξ0 ) = ηnk , which implies that ξˆnk − ξ0 ≥ 1. This is a contradiction. This proves (A.2). Suppose yn ∈ T (E), y ∈ E, yn → y. Taking a subsequence, if necessary, we may assume that yn − y < 2−n−1 ,
yn+1 − yn < 2−n .
Choose x ˆ0 such that T (ˆ x0 ) = y1 ,
ˆ x0 ≤ cy1 .
Choose x ˆn , n ≥ 1, such that T (ˆ xn ) = yn+1 − yn ,
ˆ xn ≤ cyn+1 − yn .
We set x ˆ=
∞
x ˆk .
k=0
Then we have T (ˆ x) = T
lim
n→∞
= lim
n→∞
n
xˆk
k=0
y1 +
n
k=1
= lim
n→∞
n
T (ˆ xk )
k=0
(yk+1 − yk ) = lim yn+1 = y, n→∞
which means that y ∈ T (E). Hence T (E) is closed.
Proposition A.8 Let E be a Banach space and let A : E → E be a compact operator, T = A − I. Then the equation T x = y has solutions if and only if for any solution f ∈ E ′ of the equation T ∗ (f ) = 0, one has f (y) = 0. That is, T (E) = E if and only if Ker T ∗ = {0}. Proof. (Necessity) Let x ∈ E be a solution of T x = y. Then f (y) = f (T x) = (T ∗ f )(x) = 0. (Sufficiency) Suppose y ∈ E satisfies f (y) = 0 for any solution f of the equation T ∗ (f ) = 0. If T x = y does not have solutions, then y ∈ T (E). By Proposition A7, T (E) is a closed subspace. By the Hahn-Banach theorem, there exists h ∈ E ′ such that h = 0 in T (E), h(y) = 0. Thus we have (T ∗ h)(x) = h(T x) = 0, and hence T ∗ h = 0. This contradicts the assumption.
338
Several Complex Variables and Integral Formulas
Proposition A.9 Let E be a Banach and let A : E → E be a compact operator, T = A − I. Then the equation T ∗ (f ) = g has solutions if and only if g(x) = 0 for any x ∈ Ker T . That is, T ∗ (E ′ ) = E ′ if and only if Ker T = {0}. Proof. (Necessity) Let f ∈ E ′ satisfy T ∗ (f ) = g and let x ∈ Ker T . Then we have g(x) = (T ∗ (f ))(x) = f (T x) = f (0) = 0. (Sufficiency) Suppose g ∈ E ′ satisfies g(x) = 0 for any x ∈ Ker T . Define a linear functional f0 on T (E) byf0 (y) = g(x) for y ∈ T (E), where x is one of the solutions of the equation T (x) = y. If T (x1 ) = T (x2 ) = y, then T (x1 − x2 ) = 0, which means that g(x1 ) = g(x2 + (x1 − x2 )) = g(x2 ). Hence f0 is well defined. f0 is linear since g is linear. Now we show that f0 is bounded. By (A.2) there exists a solution x ˆ of the equation T x = y such that ˆ x ≤ cy. Hence we have x)| ≤ g ˆ x ≤ cg y. |f0 (y)| = |g(ˆ Hence f0 is bounded. By the Hahn-Banach theorem, f0 is extended to a bounded linear functional F on E. Then for x ∈ E we have (T ∗ (F ))(x) = F (T x) = f0 (T x) = g(x). Hence we have T ∗ (F ) = g.
Proposition A.10 Let E be a Banach space and let A : E → E be a compact operator, T = A − I. Then T (E) = E if and only if Ker T = {0}. In this case T : E → E is surjective and invertible. Proof. (Necessity) Let Gn = Ker T n . Then Gn is a closed subspace of E and Gn ⊂ Gn+1 . Suppose T (E) = E. Assume that there exist x1 = 0 such that T (x1 ) = 0. We set T (x2 ) = x1 . Repeating this process, we have T k−1 (xk ) = x1 = 0. Since T k (xk ) = T (x1 ) = 0, we have xk ∈ Gk \Gk−1 . By Proposition A4 there exists yk ∈ Gk such that yk = 1, yk − x ≥ 1 2 (x ∈ Gk−1 ). Since {yk } is bounded, {A(yk )} contains a convergent subsequence. On the other hand, if n > m, then we have T n−1 (ym + T (yn ) − T (ym )) = T n−1 (ym ) + T n (yn ) − T n (ym ) = 0, which implies that ym + T (yn ) − T (ym ) ∈ Gn−1 . Consequently we have A(yn ) − A(ym ) = yn − (ym + T (yn ) − T (ym )) ≥
1 , 2
which contradicts {A(yk )} contains a convergent subsequence. Hence we have KerT = {0}.
Compact Operators
339
(Sufficiency) Suppose Ker T = {0}. By Proposition A.9, we have T ∗ (E ′ ) = E ′ . A∗ is a compact operator and T ∗ = A∗ − I. Using the same method as the proof of the first half, we have Ker T ∗ = {0}. By Proposition A8, we obtain T (E) = E. Finally we show that T is invertible. Since T :E → E is surjective, there is an inverse mapping T −1 : E → E. For T x = y, x satisfies (A.2), T −1 y ≤ cy, which means that T −1 is bounded. Hence T is invertible. Proposition A.11
Let X and Y be Banach spaces. Then
(a) Every compact operator T : X → Y is bounded. (b) Let T ∈ B(X, Y ) and let T (X) be a finite dimensional subspace of Y . Then T is a compact operator. (c) The set of all compact operators from X to Y is a closed subset of B(X, Y ). Proof. (a) Suppose the compact operator T : X → Y is not bounded. Then there exists {xn } such that xn = 1, T (xn ) → ∞. Since {T (xn )} does not contain any convergent subsequence, which contradicts that T is compact. (b) We denote the basis of T (X) by {e1 , · · · , ek }. Let {xn } be a bounded sequence in X. Then we have a representation T (xn ) = a1n e1 + · · · + akn ek , where {ajn }, j = 1, · · · , k, are bounded sequences. Then {a1n } contains a convergent subsequence {a1jn }. Similarly, {a2jn } contains a convergent subsequence {a2sn }. Repeating this process, {T (xn )} contains a convergent subsequence {T (xtn )}. Hence T is compact. (c) Let Tn : X → Y , n = 1, 2, · · · , be a compact operators and let T ∈ B(X, Y ), Tn − T → 0. Suppose {xn } is a bounded sequence in X. Then there exists c > 0 such that xn ≤ c. Since T1 is a compact operator, {T1 (xn )} contains a convergent subsequence {T1 (xn1 )}. Similarly, {T2 (xn1 )} contains a convergent subsequence {T2 (xn2 )}. Repeating this process, {Tk (xnn )} converges for any k. On the other hand, we have T (xmm ) − T (xnn )
≤ T (xmm ) − Tk (xmm )
+Tk (xmm ) − Tk (xnn ) + T (xnn ) − Tk (xnn )
≤ T − Tk (xmm + xnn ) + Tk (xnn ) − Tk (xmm ) ≤ 2cT − Tk + Tk (xnn ) − Tk (xmm ),
Several Complex Variables and Integral Formulas
340
which means that {T (xnn )} is a Cauchy sequence, and hence {T (xnn )} converges. Hence T is a compact operator. Proposition A.12 Let {Kn (x, y)} be a sequence of measurable functions in Ω × Ω which satisfies the following properties: (1) There exists M > 0 such that |Kn (x, y)| ≤ M (2) lim Kn (x, y) = 0 (x, y ∈ Ω).
(x, y ∈ Ω).
n→∞
For 1 ≤ p < ∞, define a linear operator Kn : Lp (Ω) → Lp (Ω) by Kn f (y) = Kn (x, y)f (x)dV (x). Ω
Then lim Kn p = 0.
n→∞
Proof.
By the H¨older inequality we have
|Kn f (y)| ≤
Ω
1/q
1/p p |Kn (x, y)|dV (x) |Kn (x, y)||f (x)| dV (x) . Ω
Consequently we have Kn f pLp
≤
M f pLp
Ω
Ω
p/q |Kn (x, y)|dV (x) dV (y).
Therefore we have limn→∞ Kn p = 0.
Proposition A.13 Let Ω ⊂ Rn be a bounded open set and let K(x, y) be a measurable function in Ω × Ω. Suppose there exists C > 0 with the properties that |K(x, y)|dV (x) ≤ C (y ∈ Ω). (1) Ω |K(x, y)|dV (y) ≤ C (x ∈ Ω). (2) Ω
For 1 ≤ p < ∞, we define a linear operator K : Lp (Ω) → Lp (Ω) by K(x, y)f (x)dV (x). Kf (y) = Ω
Then K is a compact operator.
Compact Operators
341
Proof. First we assume that K(x, y) is bounded. Then there exists C > 0 such that |K(x, y)| ≤ C. Since K is expressed by K(x, y) = K1 (x, y)+ + K1 (x, y)− + i(K2 (x, y)+ + K2 (x, y)− ), where Ki± (x, y) ≥ 0, there exists a sequence {Kn (x, y)} of simple functions which are finite linear combinations of characteristic funtions of product sets in Ω × Ω such that |Kn (x, y)| ≤ 2C and Kn (x, y) → K(k, y) in Ω × Ω almost everywhere. Since χA (x)f (x)dV (x)χB (y), χA×B (x, y)f (x)dV (x) = Ω
Ω
p
p
the range of Kn : L (Ω) → L (Ω) is a finite dimensional space. Hence By Proposition A.11 (b), Kn is compact. By Proposition A.12, K−Kn p → 0. By Proposition A.11 (c), K is compact. In the general case, we set K(x, y) (|K(x, y)| ≤ j) (j) . K (x, y) = 0 (|K(x, y)| > j) Then K (j) (x, y) is bounded, and hence compact on Lp (D) by the first part of the proof. It follows from (2) that
|K(x, y) − K (j) (x, y)||f (x)|p dV (x) dV (y) Ω Ω ≤2 |f (x)|p dV (x) |K(x, y)|dV (y) Ω
Ω
≤ 2Cf p ,
which implies that (K − K
(j)
)f pLp
≤
2Cf pLp
We set gj (y) =
Ω
Ω
Ω
|K − K
(j)
p/q |dV (x) dV (y).
|K(x, y) − K (j) (x, y)|dV (x).
It follows from (1) that |gj (y)| ≤ 2C and gj (y) → 0 (pointwise). By the Lebesgue dominated convergence theorem
p/q |K − K (j) |dV (x) dV (y) → 0 (j → ∞). Ω
Ω
By Proposition A.11 (c), K is compact.
This page intentionally left blank
Appendix B
Solutions to the Exercises
1.2 Suppose u is upper semicontinuous in Ω, that is, for any real number c, {z | u(z) < c} is an open set. For ε > 0, {z | u(z) < u(a) + ε} is an open set containing a. Hence for sufficiently small δ > 0, if |z − a| < δ, then u(z) < u(a) + ε. Consequently we have sup|z−a| 0 such that P (a, r) ⊂ Ω. It follows from Theorem 1.7 that f (ζ)dζ1 · · · dζn 1 . f (a) = · · · n (2πi) ∂P1 ∂Pn (ζ1 − a1 ) · · · (ζn − an ) Consequently, 1 (2π)n ≤ |f (a)|.
|f (a)| ≤
0
2π
···
0
2π
|f (a1 + r1 eiθ1 , · · · , an + rn eiθn )|dθ1 · · · dθn
Hence we have 2π 2π 0 / ··· |f (a)| − |f (a1 + r1 eiθ1 , · · · , an + rn eiθn )| dθ1 · · · dθn = 0. 0
0
Thus we have
|f (a)| = |f (a1 + r1 eiθ1 , · · · , an + rn eiθn )| 343
(0 ≤ θj ≤ 2π, j = 1, · · · , n),
Several Complex Variables and Integral Formulas
344
which means that |f (z)| = |f (a)| for z ∈ P (a, r). Hence |f | is constant. Since f is holomorphic in each variable, f is constant. 1.4 Since f is the limit of continuous functions which converges uniformly on every compact subset of Ω, f is continuous in Ω. Let a ∈ Ω. We choose r > 0 such that P (a, r) ⊂ Ω. It follows from Theorem 1.7 that for z ∈ P (a, r) fj (ζ)dζ1 · · · dζn 1 . (B.1) · · · fj (z) = (2πi)n ∂P1 (ζ − z1 ) · · · (ζn − zn ) 1 ∂Pn Letting j → ∞ in (B.1) we have 1 f (ζ)dζ1 · · · dζn f (z) = . · · · n (2πi) ∂P1 ∂Pn (ζ1 − z1 ) · · · (ζn − zn )
(B.2)
The right side of (B.2) can be expanded to a power series with center a (or from (B.2) one can prove ∂f /∂ z¯j = 0), which implies that f is holomorphic in P (a, r). 1.5 We choose r = (r1 , · · · , rn ) such that P (ξ, r) ⊂ Ω. It follows from Theorem 1.7 that for z ∈ P (ξ, r) f is expressed by ak1 ,··· ,kn (z1 − ξ1 )k1 · · · (zn − ξn )kn , f (z) = k1 ,··· ,kn
where ak1 ,··· ,kn =
∂ α f (ξ) α!
(α = (k1 , · · · , kn )).
Hence we have f (z) = 0 for z ∈ P (ξ, r). Since Ω is connected, we have f = 0. 1.6 Let 0 < r < 1. Define for t with −∞ < t < ∞ −1 −1 e−(t−r) e−(1−t) (r < t < 1) g(t) = . 0 (otherwise) Then g is a C ∞ function in R. We set & g A= |z1 |2 + · · · + |zn |2 dV (z), Cn & λ(z) = A−1 g |z1 |2 + · · · + |zn |2 . Then λ satisfies the following properties:
(a) λ ∈ C ∞ (Cn ).
Solutions to the Exercises
345
(b) !λ(z) = 0 for |z| ≥ 1. (c) Cn λ(z)dV (z) = 1. (d) λ depends only on |z1 |, · · · , |zn |. 1.7 Define g such that g(z) = f (z)/z for z = 0 and g(0) = f ′ (0) for z = 0. Then g is holomorphic in B(0, 1). By the maximum principle, we have |g(z)| ≤ 1. ¯1 z). 1.8 Define Φ(z) = (z + z1 )/(1 + z¯1 z) and Ψ(z) = (z − w1 )/(1 − w Then the mapping Ψ ◦ f ◦ Φ : B(0, 1) → B(0, 1) is one-to-one and onto and satisfies Ψ ◦ f ◦ Φ(0) = 0. Apply Schwarz’s lemma. 1.9 By Taylor’s formula, there exist holomorphic functions ϕ and ψ at a such that f (z) = (z − a)ϕ(z), g(z) = (z − a)ψ(z), ψ(a) = 0 Use ϕ(a) = f ′ (a) and ψ(a) = g ′ (a). 1.10 Since f (a) = 0, there exists a holomorphic function g in an open neighborhood W of a such that f (z) = (z − a)g(z) (z ∈ W ) and g(a) = 0. By the continuity, there exists an open neighborhood U ⊂ W of a such that g(z) = 0 for all z ∈ U . On the other hand, there exists a natural number N such that zn ∈ U whenever n ≥ N . Then 0 = f (zn ) = (zn − a)g(zn ) whenever n ≥ N , which is a contradiction. 1.11 Let w0 ∈ f (Ω). It is sufficient to show that f (Ω) contains an open neighborhood of w0 . There exists z0 ∈ Ω such that w0 = f (z0 ). We may assume that w0 = z0 = 0. By the uniqueness theorem (Exercise 1.10), we can choose δ > 0 such that {z | |z| ≤ δ} ⊂ Ω and f (z) = 0 for |z| = δ. We set d = min|z|=δ |f (z)|. Then d > 0. Suppose there exists w such that w ∈ f (Ω) and |w| < d. Since ϕ(z) = (f (z) − w)−1 is holomorphic in Ω, it follows from the maximum principle that 1/|w| ≤ 1/((d − |w|), which implies that {w | |w| < d/2} ⊂ f (Ω). 1.12 a ∈ Ω. Since f ′ /f is holomorphic in a simply connected domain Ω, we can define a holomorphic function ϕ in Ω such that z ′ f (ζ) dζ. ϕ(z) = a f (ζ) We set ψ = eϕ . Then a simple calculation yields d f (z) = 0. dz ψ(z) There is a constant C such that f (z) = Ceϕ(z) . Let α be an n-th root of
Several Complex Variables and Integral Formulas
346
C. Then g(z) = αe(ψ(z)/m) satisfies f = g m . Let β satisfy C = eβ . Then h = ψ + β satisfies f = eh . 1.13 Suppose there exists a ∈ Ω such that f ′ (a) = 0. By the uniqueness theorem, there exists a positive integer m (m ≥ 2) such that 0 = f ′ (a) = · · · f (m−1) (a) = 0,
f (m) (a) = 0.
Using Taylor expansion, there exists a holomorphic function g in a neighborhood of a such that f (z) = f (a) + (z − a)m g(z),
g(a) = 0.
By continuity, there exists δ > 0 such that g(z) = 0 for z ∈ B(a, δ). By Exercise 1.12, there exists a holomorphic function h in B(a, δ) such that g(z) = h(z)m for z ∈ B(a, δ). Define ϕ(z) = (z − a)h(z). Then f (z) = f (a) + ϕ(z)m
(z ∈ B(a, δ)),
ϕ(a) = 0.
Since ϕ(B(a, δ)) is an open set containing 0 by Exercise 1.11, there is ε > 0 such that {w | |w| = ε} ⊂ ϕ(B(a, δ)). Let ϕ(z0 ) = w0 and |w0 | = ε. We denote by w0 , w1 , · · · , wm−1 , the m-th roots of w0m . Then there exist z0 , z1 , · · · , zm−1 ∈ B(a, δ) such that ϕ(zi ) = wi (0 ≤ i ≤ m − 1). Therefore, f (zi ) = f (a)+w0m for i = 0, 1, · · · , m−1, which contradicts f is a one-to-one mapping. 1.14 First we show that f −1 is continuous. Suppose wn , w0 ∈ f (Ω) and wn → w0 . We set f −1 (wn ) = zn and f −1 (w0 ) = z0 . Since Ω is open, there exists r > 0 such that B(z0 , r) ⊂ Ω. For any ε > 0 with 0 < ε < r, F (B(z0 , ε)) is an open set containing w0 by Exercise 1.11. There exists an integer N such that wn ∈ f (B(z0 , ε)) whenever n ≥ N , and hence zn ∈ B(z0 , ε) whenever n ≥ N . Thus we have limn→∞ f −1 (wn ) = f −1 (w0 ). Hence f −1 is continuous. We set f (z) = w, f (z0 ) = w0 . Then if w → w0 , then z → z0 , Consequently, lim
w→w0
1 z − z0 f −1 (w) − f −1 (w0 ) = ′ . = lim z→z0 f (z) − f (z0 ) w − w0 f (z0 )
By Exercise 1.13, we have f ′ (z0 ) = 0. Hence f −1 is holomorphic. 2.2 Let K be a compact subset of Ω. For z ∈ K there exists ε(z) > 0 such that B(z, ε(z)) = {w ∈ Cn | |w − z| < ε(z)} ⊂ Ω.
Solutions to the Exercises
347
Since K is compact, by the Heine-Borel theorem there exist zi ∈ K (i = 1, · · · , p) such that p
K ⊂ ∪ B(zi , ε(zi )). i=1
We set L = ∪pi=1 B(zi , ε(zi )). Let d be the distance between K and the boundary of L. Choose ρ such that 0 < ρ < d/(3n). For z ′ , z ′′ ∈ K, |z ′ − z ′′ | < ρ, we set Γ = {w | |wj − zj′ | = 2ρ}. Then by the Cauchy integral formula we have Z fλ (ζ1 , · · · , ζn ) 1 ′ ′′ dζ1 · · · dζn fλ (z ) − fλ (z ) = (2πi)n Γ (ζ1 − z1′ ) · · · (ζn − zn′ ) Z 1 fλ (ζ1 , · · · , ζn ) − dζ1 · · · dζn . (2πi)n Γ (ζ1 − z1′′ ) · · · (ζn − zn′′ ) Since F is uniformly bounded, there exists a constant M > 0 such that |fλ (ζ)| < M
(λ ∈ Λ, ζ ∈ Ω).
Hence there exists a constant C > 0 such that |fλ (z ′ ) − fλ (z ′′ )| ≤
CM |z ′ − z ′′ | . ρ2n
Thus for any ε > 0, if we set δ = ρ2n ε(CM )−1 , then z ′ , z ′′ ∈ K, |z ′ − z ′′ | < δ ⇒ |fλ (z ′ ) − fλ (z ′′ )| < ε, which means that F is equicontinuous on K. 2.3 Let {Kj } be a sequence of compact subsets of Ω which satisfies Kj ⊂ (Kj+1 )◦ ,
∞
∪ Kj = Ω.
j=1
We choose a countable set E ⊂ Ω such that each E ∩ Kj is dense in Kj . Let E = {wi }. Since {um (w1 )} is a bounded sequence, {um } contains a subsequence {um,1} which converges at w1 . By the same reason, {um,1} contains a subsequence {um,2 } which converges at w2 . Repeating this process, there exists a subsequence {um,m } of {um } which converges pointwise in E. It follows from Exercise 2.2 that {um } is equicontinuous in Kj , which means that for ε > 0, there exists δj such that z ′ , z ′′ ∈ Kj , |z ′ − z ′′ | < δj ⇒ |um,m (z ′ ) − um,m (z ′′ )| < ε
(j = 1, 2, · · · ).
348
Several Complex Variables and Integral Formulas
Let Kj ∩ E = {ai }. Since Kj ∩ E is dense in Kj , we have ∞
Kj ⊂ ∪ B(ai , δj ). i=1
Since Kj is compact, there exists a positive integer p such that p
Kj ⊂ ∪ B(ai , δj ). i=1
(B.3)
Since {um,m } converges in Kj ∩ E, there exists a positive integer n0 such that if r, s > n0 , then |ur,r (ai ) − us,s (ai )| < ε
(i = 1, · · · , p).
Suppose z ∈ Kj . It follows from (B.3) that there exists i (1 ≤ i ≤ p) such that |z − ai | < δj . Hence, if r, s > n0 , then |ur,r (z) − us,s (z)| ≤ |ur,r (z) − ur,r (ai )| + |ur,r (ai ) − us,s (ai )| +|us,s (ai ) − us,s (z)| < 3ε, which implies that {um,m (z)} converges uniformly on Kj . Let K be an arbitrary compact subsets of Ω. Then there exists Kj such that K ⊂ Kj . Hence {um,m} converges uniformly on every compact subset of Ω. 2.4 Let b > 0. Define for x with −∞ < x < ∞ b b e− x e− a−x (0 < x < a) gb (x) = . 0 (otherwise) Then gb ∈ C ∞ (R). We set
!a gb (t)dt fb (x) = !xa . g (t)dt 0 b
Then we have fb (x) = 1 (x ≤ 0), fb (x) = 0 (x ≥ a), fb ∈ C ∞ (R), 0 ≤ fb (x) ≤ 1. Since lim gb (x) = 1 (0 < x < a), b→0
a
0
gb (x)dx → a as b → 0.
Hence if we choose b > 0 sufficiently small, then c |gb (x)| |fb′ (x)| = ! a < . a g (t)dt 0 b
fb satisfies (a), (b) and (c).
Solutions to the Exercises
349
3.1 Let x, y ∈ Γδ/2 and d = |x − y| ≤ δ/2. Then x1 +d ∂g α ′ ′ ′ |g(x1 , x ) − g(x1 + d, x )| ≤ ∂x1 (t, x ) dt ≤ C1 Kd . x1
By the mean value theorem, there exists θ such that
|g(x1 + d, x′ ) − g(y1 + d, y ′ )| ≤ Kθα−1 d, where θ is a point between x1 + d and y1 + d. Since θ > d, we have |g(x1 + d, x′ ) − g(y1 + d, y ′ )| ≤ Kdα . Then |g(x) − g(y)| ≤ |g(x1 , x′ ) − g(x1 + d, x′ )|
+|g(x1 + d, x′ ) − g(y1 + d, y ′ )|
+|g(y1 + d, y ′ ) − g(y1 , y ′ )|
≤ C2 Kdα . 3.2
dF1 (z + λθ(w − z)) |λ=1 dλ n ∂F1 ∂F1 (z + λθw)θwj + (z + λθw)θw¯j = ∂z ∂ z ¯ j j λ=1 j=1
=θ
dF1 (z + θ(w − z)) . dθ
3.3 By the Riesz representation theorem, there exists y ∈ H such that ϕ(x) = (x, y)
(x ∈ H).
For x ∈ M , we have 0 = ϕ(x) = (x, y), which implies that y ∈ M ⊥ . Suppose there exists x ∈ M ⊥ such that x, y are linearly independent. We set y y1 , y1 = x − (x, e1 )e1 , e2 = . e1 = y y1 Since {e1 , e2 } is an orthonormal system, (e1 , e2 ) = 0. Hence (y1 , y) = 0, and hence ϕ(y1 ) = 0. Thus y1 ∈ M . Since y1 is a linear combination of
Several Complex Variables and Integral Formulas
350
x and y, y1 ∈ M ⊥ . Since M ∩ M ⊥ = {0}, we have y1 = 0. Hence x and y are linearly dependent, which contradicts our assumption. Hence M ⊥ is one dimensional. 3.4 When j = k, we have (z j , z k ) =
z j z¯k dxdy =
1
2π
rj+k eiθ(j−k) rdrdθ = 0,
0
0
Ω
√ π √ . z = n+1 n
Hence {ϕn (z)} is an orthonormal sequence in A2 (Ω). A2 (Ω) → C by ψj (f ) =
∂ ∂z
j
We define ψj :
f (0)(= f (j) (0)).
Let 0 < r1 ≤ ρ ≤ r2 < 1. By the Cauchy integral formula f
(j)
j! (0) = 2πi
|z|=ρ
j! f (z) dz = z j+1 2π
2π
0
f (ρeiθ ) dθ. (ρeiθ )j
If we multiply by ρ and integrate from r1 to r2 , then we obtain f (z) j! (r2 − r1 )2 (j) f (0) = dxdy. 2 2π r1 ≤|z|≤r2 z j Consequently, we have for some constant C > 0 j! |f (j) (0)| ≤ |f (z)|dxdy π(r2 − r1 )2 r1j r1 ≤|z|≤r2 ≤ Cf . Hence ψj is a continuous linear functional. We set Mj = {f ∈ A2 (Ω) | ψj (f ) = 0}. For f ∈ Mj , we have a representation f (z) =
∞
k=0
ak z k
(ak =
f (k) (0) , |z| ≤ R < 1). k!
Solutions to the Exercises
351
Since aj = 0, we have
ϕj (z)f (z)dxdy =
B(0,R) ∞ R
=
k=0
0
0
2π
R
2π
ϕj (z)f (reiθ )rdrdθ
0
rj ei(j−k)θ ak rk+1 drdθ = 0.
0
Since R < 1 is arbitrary, (f, ϕj ) = 0, and hence ϕj ∈ Mj⊥ . Since Mj⊥ is one dimensional, Mj⊥ = {cϕj |c ∈ C}. By the Riesz representation theorem, there exists xj ∈ A2 (Ω) such that ψj (f ) = (f, xj )
(f ∈ A2 (Ω)).
We set xj = x′j + x′′j ,
(x′j ∈ Mj , x′′j ∈ Mj⊥ ).
Then we have x′′j = cj ϕj . If we set f = f1 + f2 (f1 ∈ Mj , f2 ∈ Mj⊥ ), then ψj (f ) = (f2 , x′′j ) = (f, cj ϕj ), which means that (f, ϕj ) = 0 for all j =⇒ aj = 0 for all j =⇒ f = 0. Hence {ϕn } is complete. 3.5 It follows from Exercise 3.4 that √ ∞ √ ∞ j + 1 j j + 1 ¯j 1 ¯j √ ζ = z √ KΩ (z, ζ) = (j + 1)(z ζ) π π π j=0 j=0 =
1 1 ¯ 2. π (1 − z ζ)
3.6 Assume that n = 1. (a) Let ζ ∈ Ω. For f ∈ A2 (Ω), define ψ(f ) =
∂f (ζ). ∂z
Then ψ is a continuous linear functional on A2 (Ω). By the Riesz representation theorem, there exists a function u(z, ζ) ∈ A2 (Ω) such that ψ(f ) = (f, u(z, ζ))
(f ∈ A2 (Ω)).
Several Complex Variables and Integral Formulas
352
Then u(z0 , ζ) = (u(z, ζ), KΩ (z, z0 )) = (KΩ (z, z0 ), u(z, ζ)) = ψ(KΩ (·, z0 )) =
∂KΩ (ζ, z0 ) ∂KΩ (z0 , ζ) , = ∂ζ ∂ ζ¯
which implies that ∂f (ζ) = ∂z
f (z),
∂KΩ (z, ζ) ∂ ζ¯
.
(b) We have a representation KΩ (z, z) =
∞ X
|ϕn (z)|2 ,
n=1
where {ϕj (z)} is a complete orthonormal system in A2 (Ω). If K(z, z) = 0, then for any f ∈ A2 (Ω) we have f (z) = 0. Thus KΩ (z, z) > 0. (c) We have KΩ (ζ, ζ)
∂ 2 log KΩ (ζ, ζ) ∂ 2 KΩ (ζ, ζ) = ∂ζ∂ζ ∂ζ∂ζ ∂KΩ (ζ, ζ) ∂KΩ (ζ, ζ) 1 − . KΩ (ζ, ζ) ∂ζ ∂ζ
We fix ζ ∈ Ω. We set L(z) =
∂KΩ (z, ζ) . ∂ζ
Then L ∈ A2 (Ω). We set H0 = {f ∈ A2 (Ω) | (f, KΩ (·, ζ)) = 0}. Then by the property of the Bergman kernel, H0 = {f ∈ A2 (Ω) | f (ζ) = 0}. Moreover we have Z KΩ (z, ζ)KΩ (z, ζ)dxdy kKΩ (·, ζ)k2 = (KΩ (·, ζ), KΩ (·, ζ)) = Ω Z KΩ (z, ζ)KΩ (ζ, z)dxdy = KΩ (ζ, ζ). = Ω
Now we have L(·) − αKΩ (·, ζ) ∈ H0
Solutions to the Exercises
353
⇐⇒ 0 = (L − αKΩ (·, ζ), KΩ (·, ζ)) = (L, KΩ (·, ζ)) − αKΩ (ζ, ζ) ⇐⇒ α=
(L, KΩ (·, ζ)) . KΩ (ζ, ζ)
We choose α = (L, KΩ (·, ζ))/KΩ (ζ, ζ). Then we have L − αKΩ (·, ζ)2 = (L − αKΩ (·, ζ), L − αKΩ (·, ζ)) = L2 − α(KΩ (·, ζ), L) − α ¯ (L, KΩ (·, ζ)) + |α|2 KΩ (·, ζ)2 =
=
∂KΩ (·, ζ) ∂KΩ (·, ζ) , ∂ζ ∂ζ
−
|(L, KΩ (·, ζ))|2 KΩ (ζ, ζ)
∂ 2 KΩ (ζ, ζ) ∂KΩ (ζ, ζ) ∂KΩ (ζ, ζ) 1 − . KΩ (ζ, ζ) ∂ζ ∂ζ∂ζ ∂ζ
Hence we obtain ∂ 2 log KΩ (ζ, ζ) ≥ 0. ∂ζ∂ζ Suppose there exists ζ ∈ Ω such that ∂ 2 log KΩ (ζ, ζ)/∂ζ∂ζ = 0. Then L − αKΩ (·, ζ) = 0. For f ∈ H0 we have ∂KΩ (·, ζ) ∂f (ζ). = 0 = (f, L − αKΩ (·, ζ)) = (f, L) = f, ∂ζ ∂ζ We set f (z) = z − ζ. Since Ω is bounded, f ∈ A2 (Ω) and f (ζ) = 0 which implies that f ∈ H0 . Further we have ∂f ∂ζ (ζ) = 1, which is a contradiction. 2 Hence we have ∂ log KΩ (ζ, ζ)/∂ζ∂ζ > 0. 3.7 It follows from Theorem 3.24 that Ω1 gij (z) =
∂2 log KΩ1 (z, z) ∂zi ∂ z¯j
∂2 log{|detf ′ (z)|2 KΩ2 (f (z), f (z))} ∂zi ∂ z¯j ∂2 log KΩ2 (f (z), f (z)). = ∂zi ∂ z¯j =
Several Complex Variables and Integral Formulas
354
4.1 By the expansion formula of the determinant we have ∂x ∂x1 ∂x1 ∂x1 1 ∂θ1 ∂θ2 · · · ∂θn−1 ∂r . .. .. .. .. . . . . . Jn = . ∂xn−1 ∂xn−1 ∂xn−1 ∂xn−1 ∂r ∂θ1 ∂θ2 · · · ∂θn−1 cos θ1 −r sin θ1 0 · · · 0 ∂x ∂x1 1 ∂θ1 · · · ∂θn−1 . .. .. = (−1)n+1 cos θ1 .. . . ∂xn−1 n−1 ∂θ · · · ∂x ∂θn−1 1 ∂x ∂x1 1 ∂r ∂θ2 .. .. n+2 +(−1) (−r sin θ1 ) . . ∂xn−1 ∂xn−1 ∂r ∂θ 2
We set
··· .. . ···
∂x1 ∂θn−1
.. .
∂xn−1 ∂θn−1
.
y1 = sin θ2 · · · sin θn−3 sin θn−2 sin θn−1 y2 = sin θ2 · · · sin θn−3 sin θn−2 cos θn−1 y3 = sin θ2 · · · sin θn−3 sin θn−3 cos θn−2 ··· yn−1 = cos θ2 . Then we have y1 . = (−1)n+1 r .. yn−1 = ···
∂x1 ∂θ2
.. .
∂xn−1 ∂θ2
··· .. . ···
∂x1 ∂θn−1
.. .
∂xn−1 ∂θn−1
= (−1)n+1 rn−1 sinn−2 θ1 (−1)n sinn−3 θ2 · · · (−1)5 sin2 θn−3 sin θn−1 sin θn−2 cos θn−1 × cos θn−1 − sin θn−2 sin θn−1
= ±rn−1 sinn−2 θ1 sinn−3 θ2 · · · sin2 θn−3 sin θn−2 .
Solutions to the Exercises
355
4.2 Divide the domain of integration into three parts {z | |z| < R} = {z | |z| < R, |z| < |w|/2}
∪ {z | |z| < R, |z| ≥ |w|/2, |z + w| < |w|/2}
∪ {z ||z| < R, |z| ≥ |w|/2, |z + w| ≥ |w|/2}, and use the polar coordinate system. 4.3 We set ρj (z) =
∂ρ (z), ∂zj
Φ(z, ζ) =
m j=1
ρj (ζ)(zj − ζj )
Then by Lemma 4.6, we have |Φ(z, ζ)| ≥ C(|Im Φ(z, ζ)| + |ρ(z)| +
n j=1
|ζj |2mj −2 |zj − ζj |2 + |z − ζ|M ).
For z with ρn (z) = 0, define t(ζ) = ρ(ζ) + |ρ(z)|, x2j−1 (ζ) = Re(zj − ζj ),
y(ζ) = Im Φ(z, ζ),
x2j (ζ) = Im (zj − ζj ),
j = 1, · · · , n − 1.
Then t, y, x1 , · · · , x2n−2 form a coordinate system in a neighborhood of z. Then apply the method in the proof of Theorem 3.11 and Exercise 4.2. 4.4 (1) Use the strict convexity of t2p . (2) Apply (1). 4.5 By the binomial theorem, there exist positive integers α1 , · · · , αm such that Re(z 2m ) = x2m + α1 x2m−2 (iy)2 + · · · + αm (iy)2m Then on Γσ , Re(z 2m ) ≥ x2m − α1 x2m−2 y 2 − α · · · α− my 2m
≥ x2m − α1 σ 2 x2m − · · · − αm σ 2m x2m ≥ x2m /2
for sufficiently small σ > 0. On the other hand if we choose ε > 0 sufficiently small, then ε|z|2m ≤ Re(z 2m ) on Γσ .
Several Complex Variables and Integral Formulas
356
4.6 Divide the domain of integration into 3 parts: {z | |z| < R} = {z | |z| < R, |x| < |t|/2}
∪ {z | |z| < R, |x + t| < |t|/2, |x| > |t|/2}
∪ {z | |z| < R, |x| > |t|/2, |x + t| > |t|/2}. 4.7 (1) Apply Exercise 4.4 (2) to the equation 2Re Φ(z, ζ) =
N
k=1
−γ
N
k=1
{(ηk2mk −2 − ξk2nk −2 )((xk − ξk )2 − (yk − ηk )2 ) + Re ((zk − ζk )2mk )}
= ρ(z) + +ηk2mk − N
−γ
k=1
{2nk ξk2nk −1 (xk − ξ k ) + 2mk ηk2mk −1 (yk − ηk )}
N
k {ξk2nk − x2n + 2nk ξk2nk −1 (xk − ξk ) k
k=1 yk2mk
+ 2mk ηk2mk −1 (yk − ηk )}
{(ηk2mk −2 − ξk2nk −2 )((xk − ξk )2 − (yk − ηk )2 ) + Re ((zk − ζk )2mk )}.
Then there exists δ > 0 such that 2Re Φ(z, ζ) ≤ ρ(z) − +ηk2mk −2 ((δ N
−
N
k=1
{ξk2nk −2 ((δ − γ)(xk − ξk )2 + γ(yk − ηk )2 )
− γ)(yk − ηk )2 + γ(xk − ξk )2 )}
{δ(yk − ηk )2mk + γRe ((zk − ζk )2mk )}
k=1
If 0 < γ < δ, α = min{γ, δ − γ}, then 2Re Φ(z, ζ) ≤ ρ(z) − α −
N
N
k=1
(ξk2nk −2 − ηk2mk −2 )|zk − ζk |2
{δ(yk − ηk )2mk + γRe ((zk − ζk )2mk )}.
k=1
By Exercise 4.5, if we choose γ > 0 small enough, then there exists β > 0
Solutions to the Exercises
357
such that N
k=1
{δ(yk − ηk )2mk + γRe ((zk − ζk )2mk )} ≥ β|zk − ζk |2mk .
(2) Use (1) and Exercise 4.6. 5.1 Let {zν } ⊂ ∆ be a sequence such that zν → z0 . If we set ϕjν (w) = fj (zν , w) for j = 1, 2, then ϕjν : ∆ → ∆ are holomorphic. By the Montel theorem, {ϕjν } contains a subsequence {ϕjνk } which converges uniformly on every compact subset of Ω. Let limk→∞ ϕjνk = ϕj . Since F (z, w) is a biholomorphic mapping, F (zνk , w) = (ϕ1νk (w), ϕ2νk (w)) converges to a point in ∂B. Hence (ϕ1 (w), ϕ2 (w)) ∈ ∂B which means that |ϕ1 (w)|2 + ¯ we have |ϕ′1 (w)|2 + |ϕ′2 (w)|2 = 0, Hence |ϕ2 (w)|2 = 1. Operating ∂ 2 /∂ w∂w ′ ′ we have ϕ1 = ϕ2 = 0 in ∆. Consequently we have limk→∞ Fw (zνk , w) = (ϕ′1 (w), ϕ′2 (w)) = 0. Suppose that limz→z0 Fw (z) = 0 does not hold. Then there exists a sequence {zn } and δ > 0 such that zn → z0 , |Fw (zn )| ≥ δ, which is a contradiction. This proves (a). For fixed w ∈ ∆, define Fw (z) = 0 (z ∈ ∂∆). Then by (a), Fw is continuous on ∆, holomorphic in ∆ and equals 0 on ∂∆. By the maximum principle, Fw = 0, and hence f1 (z, w) and f2 (z, w) are constant with respect to w. This proves (b). It follows from (b) that F is not one-to-one, which is a contradiction. 5.2 By the definition of the sheaf, π : S → X is a local homeomorphism. Hence there exists a neighborhood W of sx ∈ π −1 (x) such that π : W → π(W ) = U is a homeomorphism. Hence we have π −1 (x) ∩ W = {sx }. Define s = (π|W )−1 . Then we have π ◦ s(y) = y (y ∈ U ), which implies that s is a section over U . Since s(x) ∈ π −1 (x) ∩ W , we have s(x) = sx . 5.3 We set h(s) = ϕ(s)exp −
0
s
ϕ′ (θ) dθ . ϕ(θ)
Then h′ (s) = 0, and hence h is constant. Thus h(2π) = h(0) = ϕ(0). Since h(2π) = ϕ(2π)exp(−2πiN (ϕ)) = ϕ(0)exp(−2πiN (ϕ)), exp(−2πiN (ϕ)) = 1. Therefore N (ϕ) is an integer. 5.4 Let I = [0, 1]. For t ∈ I, we set ϕt (θ) = g(teiθ ). Then ϕt : [0, 2π] → C\{0} is a C 1 curve. By exercise 5.3 N (ϕt ) is an integer. Moreover ϕt (θ) and ϕ′t (θ) are continuous on [0, 2π] × I. Hence N (ϕt )
Several Complex Variables and Integral Formulas
358
is continuous with respect to t, which means that N (ϕ1 ) = N (ϕ0 ) = 0. 5.5 (a) Since Im z1 = Im z2 in A, we have for z ∈ A ∩ {z ∈ Ω | Im z1 = 0}, x2 − x1 + 1 = 0, 3/4 < |x1 | < 5/4, 3/4 < |x2 | < 5/4, which is a contradiction. (b) follows from U1 ∩ U2 = Ω\A. (c) Since f1 (1, eiθ ) = eiθ , f1 (−1, eiθ ) = eiθ + 2, we have 2π iθ 1 ie dθ = 1 N (f1 (1, eiθ )) = 2πi 0 eiθ and N (f1 (−1, eiθ )) =
1 2πi
2π
0
1 ieiθ dθ = − iθ e +2 2π
|z|=1
dz = 0. z+2
Consequently, N (f1 (−1, eiθ )) = 0 = 1 = N (f1 (1, eiθ )).
(B.4)
Suppose that f ∈ O(Ω) satisfies f /f2 ∈ O∗ (U2 ), h = f /f1 ∈ O∗ (U1 ). We set ϕt (θ) = f (eit , eiθ ) (0 ≤ θ ≤ 2π, −π ≤ t ≤ 0). Then f = f /f2 does not vanish in U2 , by the continuity of N (ϕt ) with respect to t N (f (−1, eiθ )) = N (ϕ−π ) = N (ϕ0 ) = N (f (1, eiθ )). Similarly, taking into account that h(eit , eiθ ) = 0 (0 ≤ θ ≤ 2π, 0 ≤ t ≤ π), we have N (h(−1, eiθ )) = N (h(1, eiθ )). Since f = hf1 in U1 , we obtain N (f (ζ, eiθ )) = N (h(ζ, eiθ )) + N (f1 (ζ, eiθ ))
(ζ = ±1).
Hence we have N (f1 (−1, eiθ )) = N (f1 (1, eiθ )), which contradicts (B.4).
Bibliography
[ADA1] Adachi, K. (1980). Continuation of A∞ -functions from submanifolds to strictly pseudoconvex domains, J. Math. Soc. Japan 32, pp. 331–341. [ADA2] Adachi,K. (1984). Extending bounded holomorphic functions from certain subvarieties of a weakly pseudoconvex domain, Pacific J. Math. 110, 1, pp. 9–19. [ADA3] Adachi, K. (1987). Continuation of bounded holomorphic functions from certain subvarieties to weakly psudoconvex domains, Pacific J. Math. 130, 1, pp. 1–8. [ADA4] Adachi, K. (2003). Lp extension of holomorphic functions from submanifolds to strictly pseudoconvex domains with nonsmooth boundary, Nagoya Math. J. 172, pp. 103–110. [ADC] Adachi, K., Andersson, M and Cho, H.R. (1999). Lp and H p extensions of holomorphic functions from subvarieties of analytic polyhedra, Pacific J. Math. 189, 2, pp. 201–210. [ADK] Adachi, K. and Kajimoto, H. (1993). On the extension of Lipschitz functions from boundaries of subvarieties to strongly psudoconvex domains, Pacific. J. Math. 158, 1, pp. 201–222. [ADM] Adams, R.A. (1975). Sobolev spaces, Academic Press. [AHS1] Ahern, P. and Schneider, R. (1976). The boundary behavior of Henkin’s kernel, Pacific J. Math. 66, 1, pp. 9–14. [AHS2] Ahern, P. and Schneider, R. (1979). Holomorphic Lipschitz functions in pseudoconvex domains, Amer. J. Math. 101, 3, pp. 543–565. [AMA1] Amar, E. (1983). Extension de fonctions holomorphes et courants, Bull. Sci. Math. 107, pp. 25–48. [AMA2] Amar, E. (1984). Cohomologie complexe et applications, J. London Math. Soc. 29, pp. 127–140. [BEA] Beatrous, F. (1985). Lp estimates for extensions of holomorphic functions, Michigan Math. J. 32, pp. 361–380. [BEL] Bell, S.R. and Ligocka, E. (1980). A simplification and extension of Fefferman’s theorem on biholomorphic mappings, Invent. Math. 57, pp. 283–289. [BES] Berezansky, Y.M., Sheftel, Z.G. and Us, G.F. (1996). Functional Analysis, Birkh¨ auser. 359
360
Several Complex Variables and Integral Formulas
[BR1] Berndtsson, B. (1983). A formula for interpolation and division in Cn , Math. Ann. 263, pp. 399–418. [BR2] Berndtsson, B. (1996). The extension theorem of Ohsawa-Takegoshi and the theorem of Donnelly-Fefferman, Ann. Inst. Fourier 46, pp. 1083-1094. [BRA] Berndtsson, B. and Andersson, M. (1982). Henkin-Ramirez formulas with weight factors, Ann. Inst. Fourier 32, pp. 91–110. ¨ ¨ [BRE] Bremermann, H.J. (1954). Uber die Aquivalenz der pseudokonvexen Gebiete und der Holomorphiegebiete im Raum von n komplexen Ver¨ anderlichen, Math. Ann. 128, pp. 63–91. [BRG] Bruna,J. and Burgu´es, M. (1987), Holomorphic approximation and estimates for the ∂¯ equation on strictly pseudoconvex nonsmooth domains, Duke Math. J. 55, 3, pp. 539–596. [BRJ] Bruna, J. and Castillo, J. (1984). H¨ older and Lp -estimates for the ∂¯ equation in some convex domains with real-analytic boundary, Math. Ann. 269, pp. 527–539. [BRV] Bruna, J., Cufi, J. and Verdera, J. (1985). Cauchy kernels in strictly pseudoconvex domains and an application to a Mergelyan type approximation problem, Math. Z. 189, pp. 41–53. [CH] Chen, S. C. (1990). Real analytic boundary regularity of the Cauchy transform on convex domains, Proc. Amer. Math. Soc. 108, pp. 423–432. [CHE] Chen, Z. (1990). Local real analytic boundary regularity of an integral ¯ solution operator of the ∂-equation on convex domains, Pacific J. Math. 156, pp. 97–105. [CHK] Chen, Z, Krantz, S.G. and Ma, D. (1993). Optimal Lp estimates for the ¯ ∂¯ estimates for the ∂-equation on complex ellipsoids in Cn , Manuscripta Math. 80, pp. 131–149. [CHO] Cho, H.R. (1998). A counterexample to the Lp extension of holomorphic functions from subvarieties to pseudoconvex domains, Complex Variables 35, pp. 89–91. [CUM] Cumenge, A. (1983). Extension dans des classes de Hardy de fonctions holomorphes et estimation de type ”measures de Carleson” pour l’equation ∂, Ann. Inst. Fourier 33, pp. 59–97. [DA] D’Angelo, J. (1993). Several complex variables and the geometry of real hypersurfaces, CRC Press, Inc. [DIK] Diederich, K., Fornaess, J.E. and Wiegerinck, J. (1986). Sharp H¨ older estimates for ∂¯ on ellipsoids, Manuscripta Math. 56, pp. 399–417. [DIM1] Diederich, K. and Mazzilli, E. (1997). Extension and restriction of holomorphic functions, Ann. Inst. Fourier 47, pp. 1079–1099. [DIM2] Diederich, K. and Mazzilli, E. (2001). Extension of bounded holomorphic functions in convex domains, Manuscripta Math. 105, pp. 1–12. [ELG] Elgueta, M. (1980). Extensions to strictly pseudoconvex domains of functions holomorphic in a submanifold in general position and C i nf ty up to the boundary, Ill. J. Math. 24, pp. 1–17. [FEF] Fefferman, C. (1974). The Bergmann kernel and biholomorphic mappings of pseudoconvex domains, Invent. Math. 26, pp. 1–65. [FEK] Fefferman, C. and Kohn, J.J. H (1988). ¨ older estimates on domains in two
Bibliography
361
complex dimensions and on three dimensional CR manifolds,Adv. Math. 69, pp. 233–303. [FIL] Fischer, W. and Lieb, I. (1974). Lokale Kerne und beschr¨ ankte L¨ osungen ¯ f¨ ur den ∂-Operator auf q-konvexen Gebieten, Math. Ann. 208, pp. 249–265. [FLE] Fleron, J.F. (1996). Sharp H¨ older estimates for ∂¯ on ellipsoids and their compliments via order of contact, Proc. Amer. Math. Soc. 124, pp. 3193– 3202. [FOR] Fornaess, J.E. (1976). Embedding strictly pseudoconvex domains in convex domains, Amer. J. Math. 98, pp. 529–569. [GRL] Grauert, H. and Lieb, I. (1970). Das Ramirezsche integral und die L¨ osung ¯ = α im Bereich der beschr¨ der Gleichung ∂f ankten Formen, Proc. Conf. Complex Analysis, Rice University, 56, pp. 29–50. [GRR] Grauert, H. and Remmert, R. (1979). Theory of Stein spaces, SpringerVerlag. [GUR] Gunning, R. and Rossi, H. (1965). Analytic functions of several complex variables, Prentice Hall, Englewood Cliffs, N.J. [HAT1] Hatziafratis, T.E. (1986). Integral Representation formulas on analytic varieties, Pacific J. Math. 123, 1, pp. 71–91. [HAT2] Hatziafratis, T. (1987). An explicit extension formula of bounded holomorphic functions from analytic varieties to strictly convex domains, J. Func. Anal. 70, pp. 289–303. [HEN1] Henkin, G.M. (1969). Integral representations of functions holomorphic in strictly pseudoconvex domains and some applications, Math. USSR Sb. 7, pp. 597–616. [HEN2] Henkin, G.M. (1970). Integral representations of functions in strictly ¯ pseudoconvex domains and applications to the ∂-problem, Math. USSR Sb. 11, pp. 273–281. [HEN3] Henkin, G.M. (1972). Continuation of bounded holomorphic functions from submanifolds in general position in a strictly pseudoconvex domain, Math. USSR Izv. 6, pp. 536–563. [HER] Henkin, G.M. and Leiterer, J. (1984). Theory of functions on complex manifolds, Birkh¨ auser. [HEV] Henkin, G.M. and Romanov, A.V. (1971). Exact H¨ older estimates for the ¯ solutions of the ∂-equation, Math. USSR Izvestija, 5, pp. 1180–1192. ¯ [HO1] Ho, L.H. (1991). ∂-problem on weakly q-convex domains, Math. Ann. 290, pp. 3–18. [HO2] Ho, L.H. (1993). H¨ older estimates for local solutions for ∂¯ on a class of nonpseudoconvex domains, Rocky Mount. J. Math. 23, pp. 593–607. ¯ [HR1] H¨ ormander, L. (1965). L2 estimates and existence theorems for the ∂operator, Acta Math. 113, pp. 89–152. [HR2] H¨ ormander, L. (1990). An introduction to complex analysis in several variables, Third edition, North Holland. [JK1] Jak´ obczak, P. (1983). On the regularity of extension to strictly pseudoconvex domains of functions holomorphic in a submanifold in general position, Ann. Polon. Math. 42, pp. 115–124.
362
Several Complex Variables and Integral Formulas
[JK2] Jak´ obczak, P. (1985). Extension and decomposition with local boundary regularity properties in pseudoconvex domains, Math. Z. 188, pp. 513–533. [JP] Jarnicki, M. and Pflug, R.P. (2000). Extension of holomorphic functions, De Gruyter expositions in Mathematics 34. ¯ = f in [KER] Kerzman, N. (1971). H¨ older and Lp estimates for solutions of ∂u strongly pseudoconvex domains, Comm. Pure Appl. Math. 24, pp. 301–379. [KES] Kerzman, N. and Stein, E.M. (1978).The Szeg¨ o kernel in terms of CauchyFantappi`e kernels, Duke Math. J. 45, pp. 197–224. [KON] Kohn, J.J. (1977). Methods of partial differential equations in complex analysis, Proc. symp. Pure Math. 30, pp. 215-237. [KR1] Krantz, S.G. (1976). Optimal Lipschitz and Lp regularity for the equation ¯ = f on strongly pseudo-convex domains, Math. Ann. 219, pp. 233–260. ∂u [KR2] Krantz, S.G. (1982). Function Theory of Several Complex Variables, John Wiley & Sons, New York. [KR3] Krantz, S.G. (1990). Complex Analysis: The geometric viewpoint, The Carus Mathematical Monographs 23. [KRP] Krantz, S.G. and Parks, H. (1999). The geometry of domains in space, Birkh¨ auser. [LI1] Lieb, I. (1969). Ein Approximationssatz auf streng pseudokonvex Gebieten, Math. Ann. 184, pp. 56–60. [LI2] Lieb, I. (1970). Die Cauchy-Riemannschen Differentialgleichungen auf streng pseudokonvexen Gebieten, Math. Ann. 190, pp. 6–44. [LI3] Lieb, I. (1972). Die Cauchy-Riemannschen Differentialgleichungen auf streng pseudokonvexen Gebieten: Stetige Randwerte, Math. Ann. 199, pp. 241– 256. [LIM] Lieb, I. and Michel, J. (2002). The Cauchy-Riemann Complex, Vieweg. [LIR] Lieb, I. and Range, R.M. (1980). Ein L¨ osungsoperator f¨ ur den Cauchyatzungen, Math. Ann. 253, pp. 145–164. Riemann Komplex mit C k -Absch¨ [MA] Ma, L. (1992). H¨ older and Lp estimates for the ∂-equation on non-smooth strictly q-convex domains, Manuscripta Math. 74, pp. 177–193. [MAZ1] Mazzilli, E. (1995). Extension des fonctions holomorphes, C. R. Acad. Sci. Paris 321, pp. 831–836. [MAZ2] Mazzilli, E. (1998). Extension des fonctions holomorphes dans les pseudo ellipso¨ıdes, Math. Z. 227, pp. 607–622. [MEN] Menini, C. (1997). Estimations pour la resolution du ∂¯ sur une intersection d’ouverts strictement pseudoconvexes, Math. Z. 225, pp. 87–93. ¯ [MIC] Michel, J. (1988). Randregularit¨ at des ∂-Problems f¨ ur st¨ uckweise streng pseudokonvexe Gebiete in Cn , Math. Ann. 280, pp. 45–68. ¯ [MIP] Michel, J. and Perotti, A. (1990). C k -Regularity for the ∂-equation on strictly pseudoconvex domains with piecewise smooth boundaries, Math. Z. 203, pp. 415–427. [NOR] Norguet, F. (1954). Sur les domains d’holomorphie des fonctions uniformes de plusieurs variable complexes(passage du local au global), Bull. Soc. Math. France 82, pp. 137–159. [OHT] Ohsawa, T. and Takegoshi, K. (1987). On the extension of L2 holomorphic functions, Math. Z. 195, pp. 197–204.
Bibliography
363
[OKA1] Oka, K. (1942). Sur les fonctions de plusieurs variables, VI. Domaines pseudoconvexes, Tohoku Math. J. 49, pp. 15–52. [OKA2] Oka, K. (1950). Sur les fonctions de plusieurs variables, VII. Sur quelques notions arithm´etiques, Bull. Soc. Math. France 78, pp. 1–27. ˙ [OKA3] Oka, K. (1953). Sur les fonctions de plusieurs variables, Domaines finis sans point critique int´eerieur, Jap. J. Math. 23, pp. 97–155. [OV] Ovrelid, N. (1971). Integral representation formulas and Lp estimates for the bar∂-equation, Math. Scand. 29, pp. 137–160. [POL] Polking, J. (1991). The Cauchy Riemann equations on convex sets, Proc. Symp. Pure Math. 52, pp. 309–322. [RAM] Ramirez, E. (1970). Ein Divisionsproblem und Randintegraldarstellungen in der komplexen Analysis, Math. Ann. 184, pp. 172–187. ¯ = f on weakly pseudo[RAN1] Range, R.M. (1976). On H¨ older estimates for ∂u convex domains, Int. Conf. Cortona. [RAN2] Range, R.M. (1986). Holomorphic functions and integral representations in several complex variables, Springer. [RAN3] Range, R.M. (1990). Integral kernels and H¨ older estimates for ∂¯ on pseudo-convex domains of finite type in C2 , Math. Ann. 288, pp. 63–74. [RAN4] Range, R.M. (1992). On H¨ older and BMO estimates for ∂¯ on convex domains in C2 , J. Geom. Anal. 2, pp. 575–583. ¯ [RAS] Range, R.M. and Siu. Y.T. (1973). Uniform estimates for the ∂-equation on domains with piecewise smooth strictly pseudoconvex boundaries, Math. Ann. 206, pp. 325–354. ¯ [SCH] Schmalz, G. (1989). Solution of the ∂-equation with uniform estimates on strictly q-convex domains with nonsmooth boundary, Math. Z. 202, pp. 409–430. [SI1] Siu, Y.T. (1974). The ∂¯ problem with uniform bounds on derivatives, Math. Ann. 207, pp. 163–176. [SI2] Siu, Y.T. (1996). The Fujita conjecture and the extension theorem of Ohsawa-Takegoshi, Proc. 3rd Int. RIMSJ., Geometric Complex Analysis, pp. 577–592. [STE] Stein, E.M. (1972). Boundary behavior of holomorphic functions of several complex variables, Math. Notes ♯11, Princeton Univ. Press. [STO] Stout, E.L. (1975). An integral formula for holomorphic functions on strictly pseudoconvex hypersurfaces, Duke Math. J. 42, pp. 347–356. [TSU] Tsuji, M. (1995). Counterexample of an unbounded domain for Ohsawa’s problem, Complex variables 27, pp. 335–338. [VER] Verdera, J. (1984). L∞ -continuity of Henkin operators solving ∂¯ in certain weakly pseudoconvex domains of C2 , Proc. Royal Soc. Edinburgh 99 , pp. 25–33.
This page intentionally left blank
Index
Cauchy-Fantappi`e formula, 132 closed operator, 53 coboundary, 317 cochain, 317 cocycle, 317 coherent sheaf, 316 cohomology group, 317 compact operator, 227, 331 complete orthonormal system, 198 condition (Bk ), 237 condition (R), 230 condition (Rk ), 230 conjugate operator, 331 continuous extension, 181 convex domain of finite type, 288 convex hull, 23
C ∞ extension, 196 C k extensions, 189 H p extension, 188 L2 estimate for the ∂¯ problem, 91 Lp estimates for the ∂¯ problem, 263 P S(Ω)-convex, 33 X intersects ∂Ω transversally, 158 O(Ω)-convex, 23 admissible kernel, 217 analytic polyhedron, 43 analytic sheaf, 308 analytic subset, 320 Ascoli-Arzela theorem, 331 Baire’s theorem, 14 Banach-Steinhaus theorem, 56 Bell’s density lemma, 233 Bergman kernel, 205 Bergman metric, 242 Bergman projection, 228 Bergman space, 202 Berndtsson formula, 265 Berndtsson-Andersson formula, 254 Bessel’s inequality, 197 Bochner-Martinelli formula, 125 bounded extension, 181
Dini’s theorem, 36 domain of convergence, 10 domain of holomorphy, 22 E. M. Stein’s counterexample, 150 Fefferman’s mapping theorem, 241 first Cousin problem, 326 frame, 43 function element, 307
Carath´eodory metric, 291 Cartan theorem A, 321 Cartan theorem B, 321 Cauchy inequality, 13
germ, 307 Gram-Schmidt orthonormalization process, 196 Green’s theorem, 48 365
366
Several Complex Variables and Integral Formulas
H¨ older estimate for the ∂¯ problem, 148 H¨ older space, 123 Hahn-Banach theorem, 56 Hartogs theorem, 16 Henkin-Ramirez kernel, 148 holomorphic, 11 holomorphically convex, 23 holomorphically convex hull, 23 homotopy formula, 132 implicit function theorem, 302 inverse mapping theorem, 304 invertible operator, 331 kernel, 60 Kobayashi metric, 292 Koppelman formula, 127 Koppelman-Leray formula, 132 Leray formula, 131 Leray map, 129 Levi polynomial, 143 Levi’s problem, 95 Lipschitz space, 123 locally finitely generated, 315 maximum principle, 44 Montel’s theorem, 112 Narashimhan’s lemma, 136 Noetherian ring, 309 nonsingular mapping, 303 null space, 60 Ohsawa-Takegoshi theorem, 114 Oka’s counterexample, 330 Oka’s fundamental theorem, 311 open mapping theorem, 46 outward normal vector, 48 Parseval’s equality, 199 peak function, 190 plurisubharmonic, 19 Poincar´e theorem, 297, 329
Poisson integral, 5 pseudoconvex open set, 32 pullback, 118 regular of order k, 298 relatively compact, 11 Riesz representation theorem, 56 Riesz-Fischer theorem, 199 Schwarz lemma, 45 Schwarz-Pick lemma, 45 second Cousin problem, 328 section, 308 separable, 86 sheaf, 308 sheaf homomorphism, 308 sheaf of Abelian groups, 308 sheaf of ideals, 320 sheaf of modules, 308 sheaf of relations, 316 sheaf of rings, 308 simple admissible kernel, 217 Sobolev space, 81 stalk, 308 Stein cover, 322 Stein manifold, 319 strictly plurisubharmonic, 19 strictly pseudoconvex domain, 33 strictly subharmonic, 4 subharmonic, 3 submanifold in general position, 159 subsheaf, 308 surface element, 48 surface measure, 149 uniqueness theorem, 45 upper semicontinuous, 3 weak convergence, 86 Weierstrass division theorem, 300 Weierstrass polynomial, 299 Weierstrass preparation theorem, 299